This is the full developer documentation for Supervertaler Docs # Welcome to Supervertaler Docs > Supervertaler for Trados brings AI translation, live terminology, cross-file search and a translation knowledge base into Trados Studio. Supervertaler for Trados  A plugin that brings Supervertaler’s AI and terminology tools right into [Trados Studio](https://www.trados.com/) (2024 and 2026) as dockable panels – so you never leave the CAT tool you already use. What it adds: * **TermLens** – live, colour-coded terminology under each source segment, with MultiTerm (`.sdltb` / `.ttb`) support and one-key term insertion (`Alt+0`–`9`) * **Supervertaler Assistant** – a Trados-aware AI chat that knows your current segment, terminology, and TM matches; attach files, go incognito, or draw on its **SuperMemory** knowledge base * **SuperSearch** – cross-file search and replace across every file in the project * **Batch Translate & AI Proofreader** – run AI over entire files, with a prompt library you can tailor to your domain * **QuickLauncher** – one-keystroke prompt actions, hands-free * **MCP Server** – connect Claude Desktop and other AI assistants to the live Studio session Source-available, single paid plan, 14-day free trial. [Open the docs →](/trados/) Supervertaler for memoQ  A plugin for [memoQ](https://www.memoq.com/) that adds an AI translation engine which learns from the segments you confirm, and puts your own glossary into both memoQ’s terminology pane and the AI’s prompt. * **AI translation** – Anthropic, OpenAI or Google, with your own API key, segment by segment or through Pre-translate * **Self-learning** – confirm a term once and the rest of the document follows it * **Terminology** – your glossary highlighted in the grid *and* enforced in the prompt, forbidden terms included **In early development.** [Open the memoQ docs →](/memoq/) Supervertaler Workbench 🖥️ A free, open-source standalone CAT tool – editor, AI translation, terminology and translation memory in one place, plus tools that work system-wide in any application: a clipboard manager, SuperLookup, QuickTrans, and voice dictation. **No longer actively developed.** It still works and the source is still open, but it is not receiving new features. Its documentation stays online in full. [Open the Workbench docs →](/workbench/) # Supervertaler for memoQ Supervertaler for memoQ brings AI translation and your own terminology into **memoQ 12**, as two add-ins that work together. Unlike the Trados plugin, which docks its own panels into the editor, memoQ gives an add-in no window of its own. Supervertaler therefore works through memoQ’s existing surfaces – the machine-translation engine and the terminology pane – rather than adding new ones. In practice that turns out to suit it: the results appear exactly where you already look for them. ### What it does **AI translation.** An LLM machine-translation engine using Anthropic, OpenAI or Google, with your own API key and your own instructions. It works segment by segment as you translate, and in bulk through **Pre-translate**. Inline tags survive the round trip. **It learns as you work.** Every segment you confirm is remembered, and the most relevant ones are shown to the model when it translates later segments in the same document. Settle on a term once and the rest of the document follows it – no configuration, no retraining, just your own approved choices fed forward. See [Self-learning translation](/memoq/self-learning/). **Your terminology, twice over.** A glossary you point Supervertaler at appears as a memoQ terminology provider – matched terms highlighted in the source, entries listed in Translation results – *and* is sent to the model as required or forbidden terminology. Forbidden terms are enforced, not merely displayed. See [Terminology](/memoq/terminology/). **Translate with Claude Desktop.** Through the [Supervertaler MCP Server](/memoq/mcp-server/), Claude reads the document you are translating, your confirmed segments and your glossary, and stages translations that flow into the grid when you press Pre-translate. Tokens are billed to your Claude subscription rather than an API key, and every write into your document goes through your own hands. See [MCP Server](/memoq/mcp-server/). **A prompt library, shared with Trados.** Translation instructions come from the same library the Trados plugin uses, chosen from a dropdown and edited in a small companion [editor](/memoq/prompt-editor/). Claude can draft prompts into it too. ### What it does not do memoQ does not let a plugin read its own term bases or translation memories, so terms defined in a memoQ term base are not visible to the AI. Supervertaler reads its own glossary file instead, which it can also display alongside memoQ’s own term base hits. There is no chat panel, no document-wide search and no cursor control inside memoQ: a plugin can answer when asked for a translation, and that is all. The chat lives in Claude Desktop; the prompt library lives in its own editor; and Claude’s translations reach the grid only when you Pre-translate. [MCP Server → What it can and cannot do](/memoq/mcp-server/#what-it-can-and-cannot-do) has the full comparison with the Trados plugin. ### Where to start * [Installation](/memoq/installation/) – putting the add-ins in place * [Getting started](/memoq/getting-started/) – a first translation * [Terminology](/memoq/terminology/) – using a glossary * [MCP Server](/memoq/mcp-server/) – translating with Claude Desktop * [Prompt Library & Editor](/memoq/prompt-editor/) – choosing and writing instructions # Context layers > The layers of context Supervertaler for memoQ puts in front of the AI, what each one adds, where memoQ lets it come from, and how to control them. A translation engine that sees only the sentence in front of it will translate that sentence well and the document badly. Supervertaler’s design is the opposite: every time memoQ asks it for a translation, it assembles a fresh snapshot of your work and hands the whole thing to the AI. That snapshot is built in **layers**. Some come from memoQ, some from your own recorded knowledge, and two of them come from parts of the document that no CAT tool normally shows you at all. They stack, and each one that is present removes a class of mistake the AI would otherwise make. This page is the single place that lists every layer. Supervertaler for Trados has [its own version of this page](/trados/context-layers/); the ideas are the same and the sources differ, because memoQ hands a plugin a different set of things. ## The layers | # | Layer | Where it comes from | Needs work from you | | -- | ------------------------------------ | --------------------------------------- | ------------------------ | | 1 | Project and job information | memoQ, with each request | no | | 2 | The segment being translated | memoQ | no | | 3 | Segments you have confirmed | your own confirmed rows | one setting, once | | 4 | The closest translation memory match | your TMs, routed by you | one setting, once | | 5 | Whether the row was rejected | memoQ | no | | 6 | Glossary terms and forbidden terms | your glossary | a glossary | | 7 | SuperMemory memory banks | what you have recorded about the client | a bank per client | | 8 | List numbering | the original Word file | the live document link | | 9 | Figure descriptions | the images in your documents | two clicks in FigureLens | | 10 | The whole document | the live document link | for AutoPrompt only | ### 1. Project and job information The client, domain and subject recorded in memoQ’s project, the language pair, and which document the segment is in. memoQ sends these with every translation request, so they need no setting – but they are only as good as what the project manager filled in. An empty *Client* field in memoQ is an empty line in the prompt. **Toggle:** Translation settings → *Send surrounding segments and project metadata to the model*. ### 2. The segment being translated The source text, with its inline tags preserved so they can be put back in the right places. Always included; there is nothing to configure. ### 3. Segments you have confirmed Every segment you confirm is recorded, and the most lexically similar ones are shown to the model when it translates later segments of the same document. Confirm *electric module* once and the rest of the document follows. This is memoQ’s answer to a layer Trados gets for free. A Trados plugin is handed the segments surrounding the one it is translating; a memoQ MT plugin is not – memoQ reserves that channel for its own AGT engine – so Supervertaler builds the equivalent out of your confirmed work instead. It is arguably the better trade: every example is one you approved, rather than merely one that happens to be nearby. memoQ only sends confirmations to an engine selected under **Self-learning MT**, so that box must be ticked as well as the engine being chosen for translation. Until it is, nothing is captured and the [Activity window](/memoq/prompt-editor/#the-activity-window) says so. See [Self-learning](/memoq/self-learning/). **Toggle:** Translation settings → *Send surrounding segments and project metadata to the model*. ### 4. The closest translation memory match memoQ can forward the best fuzzy match for a segment to an MT engine, and when you route it to Supervertaler that match goes into the prompt ahead of everything else – presented as the thing to adapt rather than as background reading, because a human wrote and approved it for a nearly identical source. Set it in memoQ under **Edit machine translation settings → Send best fuzzy TM match to → Supervertaler**. It is deliberately *not* governed by the document-context toggle: a match you went out of your way to route here should not disappear because you turned off surrounding context. Two things to know about its shape. It is **one** match per segment – memoQ forwards the single best one, which is what its own setting says – rather than a set to choose among. And memoQ hands over the two segments without a match rate, so the prompt cannot tell the model *how* close the match is; it is described as the closest approved rendering, and the model judges the difference from the text itself. The match is forwarded for every segment memoQ asks about, batches included. ### 5. Whether the row was rejected memoQ tells the plugin each row’s translation state. When you have rejected a previous translation of a segment, the prompt says so and instructs the model to reconsider the terminology, structure and register rather than paraphrase what you refused. No setting; it simply happens on a row you marked rejected. ### 6. Glossary terms and forbidden terms Terms matched in the segment, with their approved renderings, and any terms marked forbidden. Forbidden terms are enforced rather than merely displayed. Worth being precise about the source, because the setting’s wording is optimistic: these come from **Supervertaler’s own glossary** – the tab-separated file the [terminology plugin](/memoq/terminology/) reads – not from memoQ’s term bases. memoQ passes termbase hits to an MT plugin only through the rich lookup channel it reserves for its own engine, so a third-party plugin never receives them. Your memoQ term bases still work normally in the grid; they just do not reach the model. [Export glossary](/memoq/prompt-editor/#export-glossary-the-prompts-terms-as-the-project-glossary) is the bridge: it turns a drafted prompt’s locked terms into a glossary Supervertaler does read. While an AutoPrompt-drafted prompt is selected, its own locked-terms table is the authority and the glossary’s preferred renderings are held back – forbidden terms always travel. **Toggle:** Translation settings → *Send memoQ’s termbase hits and forbidden terms to the model*. ### 7. SuperMemory memory banks A bank of Markdown articles – `brief.md`, `terminology.md`, `style.md`, and any others you add – goes out with every request, up to about 32,000 tokens, and to AutoPrompt up to 40,000. Where a glossary gives the model flat pairs of terms, a memory bank gives it the **reasoning**: the decisions, the caveats, the client-specific overrides. The `_shared` bank travels alongside as house defaults. Banks are remembered per memoQ project, and a project you have never chosen one for uses none rather than inheriting the last – a bank carries one client’s terminology, and the wrong one is worse than none. See [Memory banks](/memoq/mcp-server/#memory-banks). ### 8. List numbering Word numbers claims, letters steps and bullets lists as paragraph properties, not as text, so memoQ’s grid never contains the `a)` or the `9.` and neither did anything the model received. Shown six unlettered steps and then *“steps a. to f.”*, a model will flag the reference as a possible defect in the source – a note that would have reached the client. Supervertaler reads the numbering out of the original `.docx`, counted over the whole document exactly as Word renders it, and sends each paragraph’s marker in front of its first segment as `[#e)]`, declared as structure to use and never to reproduce. Anything echoed back is stripped before it reaches the document. On by default, with no switch – but it needs the [live document link](/memoq/mcp-server/#the-live-document-link) connected, because that is what names the file memoQ imported. On a project checked out from a server that file is on the project manager’s machine, not yours; locate it once in [FigureLens](/memoq/prompt-editor/#where-the-documents-come-from) and the numbering is read from your copy. See [List numbering](/memoq/prompt-editor/#list-numbering-reaches-the-model-as-structure). ### 9. Figure descriptions The AI reads your text and cannot see your pictures. [**FigureLens**](/memoq/prompt-editor/#figurelens-what-the-figures-show) closes that gap: it takes the images out of your documents into the memory bank’s `figures\` folder, shows each one to the AI together with what the document says about it, and saves a description – what it shows, which figure it is, which reference signs appear on it – as `figures.md` in the bank. From then on it rides along with layer 7 and is read with every request, so the model knows that the *valve (12)* in the sentence is the thing at the top right of Figure 3. Vector drawings – EMF and WMF, which is what a drawing placed from CAD usually is – are rendered to PNG on the way out, because no AI can read a metafile. Reference signs the model reads in a drawing that appear nowhere in the text are listed for you, because that is a defect worth raising with the client before filing. Two clicks per project, then it is automatic. One AI request per image; the panel says how many and to which provider before you click, and **Describe from the text only** is a free alternative that uses what the document itself says about each figure. ### 10. The whole document Not a translation layer: memoQ hands an MT plugin about ten segments at a time and never the document, so a per-segment prompt cannot contain it. What does get the whole document is [**AutoPrompt**](/memoq/prompt-editor/#autoprompt-drafting-a-prompt-for-the-open-project), which reads it through the live document link to classify the job and draft a prompt for it. That prompt then carries the document’s character into every subsequent request – which is the point: the document is read once, expensively, and its conclusions ride along cheaply thereafter. Where memoQ shows several files merged into one view, AutoPrompt offers **All N documents memoQ is showing** so the prompt is drafted from all of them rather than whichever file you happened to pick. ## What memoQ does not give a plugin Worth stating plainly, because the Trados page lists them and the difference is not a fault in either product: * **Surrounding segments.** memoQ passes neighbouring segments only through the rich lookup channel reserved for its own AGT engine. Layer 3 exists because of this. * **Term base hits.** The same channel, the same result. See layer 6. * **Attached files.** memoQ has no equivalent; there is nothing to attach a file to. None of these can be unlocked by a setting on your side. They are consequences of the plugin model, measured rather than assumed – and where one bites, the layer above it says what stands in for it. ## Stacking the layers The default composition – everything above except the ones that need a click – is a strong baseline and the one to start from. More context is not automatically better. The context window is finite, and a mature memory bank plus a rich glossary can push a prompt into the tens of thousands of tokens. The cost is smaller than it looks, because the stable half of every request is identical from batch to batch and is cached: on one 370-segment run, caching turned roughly $10 into roughly $4. But token count is not the only cost – three overlapping sources describing the same term can contradict each other, and the model then has to reconcile them on the fly. The layers that describe the **document** rather than the **client** – 5, 8 and 9 – are cheap, never contradict each other, and there is no case yet found where switching one off improved a translation. ## Seeing what is being sent The [Activity window](/memoq/prompt-editor/#the-activity-window) in the prompt editor logs every request with its token counts – regular, cached, written, output – so a layer that is not arriving shows up as a number that does not move. It also says, once per document, whether list numbering was available and why not when it was not, and which memory bank is in force. AutoPrompt’s **Preview context…** button shows exactly what would be sent before anything is sent, and makes no AI call. ## See also * [Prompt library and editor](/memoq/prompt-editor/) – AutoPrompt, FigureLens, list numbering, the settings * [Self-learning](/memoq/self-learning/) – how confirmed segments are captured and fed back * [Terminology](/memoq/terminology/) – the glossary the model reads * [MCP server and the live document link](/memoq/mcp-server/) – memory banks, and the channel layers 8 to 10 depend on * [Context layers in Supervertaler for Trados](/trados/context-layers/) – the same idea, a different host # Getting Started This walks through a first translation, assuming the add-ins are [installed](/memoq/installation/). ### 1. Set up the translation engine Open the **Resource console** → **MT settings**. Create or edit an MT settings resource, and on the **Services** tab tick **Supervertaler**. Click **Configure plugin** and fill in: | | | | ------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Provider** | Anthropic, OpenAI or Google | | **Model** | a short list of the models worth recommending for that provider, with a line on each saying what it is for. **Fetch list** asks the provider for its full catalogue and **Show all models** adds it underneath – that is where to look for a model released after your copy of Supervertaler was built. The box also takes anything typed, for a gateway or a local model | | **API key** | your own key for that provider. All Supervertaler products share one key file at `C:\Users\\Supervertaler\settings\api-keys.json`, so a key you have already set in Supervertaler for Trados or Sidekick is picked up here and you can leave this empty – see [Prompt Library & Editor](/memoq/prompt-editor/#api-keys-live-in-one-file) | | **Endpoint** | leave blank unless you are using a local model or a gateway | | **Segments per request** | how many segments go into one request during Pre-translate (memoQ hands the plugin about 10 at a time, so values above 10 make no difference) | | **Prompt** | a prompt from the shared library, or *(instructions below)* to type your own – see [Prompt Library & Editor](/memoq/prompt-editor/) | | **Pre-translate via Claude Desktop (MCP)** | leave **unticked** unless you translate through Claude Desktop – see [MCP Server](/memoq/mcp-server/). Also in the prompt editor under Settings | Press **Test connection**. It translates a short sentence for real, so it exercises the key, the model name and the endpoint together – a green result means everything works. ### 2. Turn on learning Still in the MT settings resource, go to the **Settings** tab and set **Self-learning MT** to **Supervertaler**. This is what makes memoQ hand Supervertaler each segment as you confirm it. Without it the engine still translates, but it will not learn from your work. See [Self-learning translation](/memoq/self-learning/). Caution Changing this takes effect when memoQ next builds a translation engine. Restart memoQ after setting it. ### 3. Translate Open a document. With **Translation results** set to *Always*, landing on a segment fetches a translation automatically; it appears in the Translation results pane, labelled with the provider and model it came from. To translate in bulk, use **Preparation → Pre-translate** with *Use machine translation* enabled. ### 4. Confirm as you go Confirm segments as you normally would. Each confirmation is remembered, and later segments that resemble it are translated with your wording in front of the model. The effect is most visible on a document with recurring phrasing: settle a term in segment 3 and segment 40 will use it. ### Next * [Terminology](/memoq/terminology/) – add a glossary, including forbidden terms * [Self-learning translation](/memoq/self-learning/) – what is remembered, and for how long # Glossary Format A plain text file, tab-separated, one term per line. ```plaintext elektrische module electric module elektrische module electrical module forbidden koppelmechanisme coupling mechanism ``` | Column | | | ------ | ------------------------------------------------------------- | | 1 | source term | | 2 | target term | | 3 | *optional* – `forbidden` marks a target that must not be used | Blank lines are ignored, and so is any line starting with `#`, so the file can carry comments. Tabs, not spaces Columns must be separated by actual tab characters. This is the commonest reason a glossary loads with no terms. The options dialog reports how many terms it parsed – check that number after choosing a file. ### Matching Matching is case-insensitive and respects word boundaries, so `wire` does not match inside `wireless`. Where two entries could both match, the longer wins and the shorter one inside it is suppressed: `electric module` beats a bare `module`. ### Editing while you work The file is re-read whenever you save it. Keep it open in a text editor beside memoQ, add a term, save, and the next segment sees it – no restart, no reloading the project. ### Size A real term base export works fine. A 9,000-term glossary loads in well under a tenth of a second and adds under a millisecond to each lookup. Because matching is case-insensitive, very short entries are worth avoiding: a two-letter term fires on almost every segment and crowds out terms that matter. ### Converting an existing term base The plugin repository includes a converter for Supervertaler Workbench term base exports: ```bash python tools/convert_termbase.py BEIJER.tsv BEIJER-glossary.txt ``` It handles the quoting and pipe-separated variants of the export format, drops entries that translate to themselves, and reports what it skipped. # Installation ### Requirements * **memoQ 12** (translator pro or project manager) * An API key for Anthropic, OpenAI or Google * Administrator rights on the machine, once, to place the files ### Installing Supervertaler for memoQ ships as two files: ```plaintext Supervertaler.MemoQ.dll the AI translation engine Supervertaler.MemoQ.Terms.dll the terminology provider ``` Both belong in memoQ’s `Addins` folder, inside the memoQ program directory – typically: ```plaintext C:\Program Files\memoQ\memoQ-12\Addins\ ``` Copy both files there and restart memoQ. Why two files memoQ loads exactly one add-in per DLL, so the translation engine and the terminology provider cannot share one. They are halves of the same product and are designed to be installed together – but the terminology half is optional if you only want AI translation. ### The unsigned plugin warning The first time memoQ starts after installation it will say it has detected one or more unsigned plugins, and ask whether to load them. **Answer Yes.** The default button is *No*, so a stray Enter keypress will decline it. Supervertaler is not yet signed by memoQ. Signing is a review process memoQ runs for plugins with proven demand; until then, this prompt appears whenever the files change. ### Where the program directory is version-stamped memoQ’s install folder carries its version number (`memoQ-12`, `memoQ-13`, …). A memoQ major upgrade creates a **new folder**, and the add-ins are not carried across – they will need to be copied again. If Supervertaler disappears after a memoQ update, this is almost always why. ### Uninstalling Close memoQ, delete the two DLLs from the `Addins` folder, and restart. To remove what Supervertaler has stored on your computer as well, use **Forget stored context** in the plugin’s options dialog before uninstalling – see [Self-learning translation](/memoq/self-learning/#what-is-stored-and-where). # MCP Server (Claude Desktop) Supervertaler for memoQ can connect **Claude Desktop** – or any AI app that runs a local MCP server – to your live memoQ project. You chat in Claude’s window; Claude reads the document you are translating, your confirmed segments and your glossary, and translates for you. The tokens are billed to your Claude subscription, not to an API key. It is the same [Supervertaler MCP Server](/trados/mcp-server/) the Trados plugin uses. What differs is what memoQ lets a plugin do – which is a good deal less than Trados – so read [What it can and cannot do](#what-it-can-and-cannot-do) before you expect Trados behaviour. > **Which AI apps work?** The same answer as for Trados: any app that runs a **local (STDIO) MCP server on your own machine** – Claude Desktop, ChatGPT’s desktop app, Claude Code. Cloud-hosted clients (the claude.ai and chatgpt.com websites) have no route to a bridge that lives on your PC. ## The one thing to understand first **memoQ never lets a plugin write into the grid.** A plugin cannot move your cursor, edit a segment or confirm anything. It can only *answer when memoQ asks it for a translation*. So Claude does not write translations into memoQ. It **stages** them. They wait inside the plugin until you run **Pre-translate** (or land on the segment), at which point memoQ asks Supervertaler for a translation and receives Claude’s. Every write into your document goes through your own hands – which is not a limitation so much as a built-in review step. The other half: a plugin only *sees* what memoQ sends it. Claude cannot read your document until Supervertaler has been shown it. One Pre-translate pass does that. ## The workflow With everything set up (below), a chat-driven job looks like this: 1. **Open the project** in memoQ and tick **Pre-translate via Claude Desktop (MCP)** in Supervertaler’s settings (see [The checkbox](#the-checkbox)). 2. **Pre-translate** with Supervertaler as the MT engine. It is instant and free: the grid stays empty, but Supervertaler now holds every source segment. 3. **In Claude Desktop:** *“Read my memoQ project and translate it into Dutch.”* Claude reads the segments, checks your glossary, and stages translations. Nothing has changed in memoQ yet. 4. **Pre-translate again.** The grid fills. Each row is marked `Claude (staged via Supervertaler MCP)` in Translation results. 5. **Confirm as you go.** If [Self-learning](/memoq/self-learning/) is on, each confirmation is visible to Claude too – *“what have I confirmed so far?”* – so a mid-job conversation about terminology is grounded in your actual choices. Rows Claude has not staged still get a live suggestion from the model as you land on them, exactly as before. The checkbox only changes what **Pre-translate** does. ## The checkbox Either in the [prompt editor](/memoq/prompt-editor/), on the **Settings** menu, which needs no project open and no memoQ running, or in memoQ under **Resource console → MT settings → Supervertaler → Configure plugin**: > ☐ **Pre-translate via Claude Desktop (MCP) instead of the API key above** *Pre-translate then only hands the segments to the chat and inserts the translations it sends back; nothing is charged to the API key. Suggestions as you move through segments still use the API key.* Both paths call an AI model. The checkbox decides **which one pays and who drives**: unticked, this plugin translates through the API key you entered; ticked, Pre-translate leaves the translating to the chat app, billed to that subscription. | | Pre-translate | Landing on a segment | | ---------------------- | --------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------- | | **Unticked** (default) | This plugin translates every segment through your API key. Staged translations are used first where they exist. | Live suggestion from the model; staged first if present. | | **Ticked** | Hands the segments to the chat and inserts what it sends back. Nothing charged to the API key. | Unchanged – live suggestion from the model. | Two consequences worth knowing: * **You never have to toggle it mid-job.** Ticked is right for the whole of a chat-driven job: the capture pass is free, the delivery pass is free, and walking the document afterwards still gives you live suggestions for anything Claude did not cover. * **It is one switch for the whole installation, not a project setting.** It says how you are working at this moment, so the editor and memoQ show the same state and either can change it. * **Staged translations come through in either state.** Ticking the box is never a way to lose Claude’s work; unticking it is never a way to block it. It only decides whether *Pre-translate* spends API money on rows nothing was staged for. Leave it unticked if you use Supervertaler as an ordinary MT engine with no chat involved. That is the default, and it is what most memoQ users will want. ## Setting it up **Claude Desktop:** install the extension. 1. Download `Supervertaler-for-memoQ-MCP-Server.mcpb` (it ships with the plugin). 2. In Claude Desktop, open **Settings → Extensions → Advanced settings** and click **Install extension…** (double-clicking the file also works if `.mcpb` is associated with Claude; drag-and-drop does not). 3. In memoQ, open a project and click into any segment with Supervertaler selected as the MT engine. That creates the engine, which starts the bridge. 4. In Claude: *“What’s in my memoQ project?”* If it answers with your language pair and segment count, you are connected. If you also use the Trados plugin, both extensions coexist – Claude shows them as two servers – and they are in fact the same server exe: the memoQ one carries a single setting, `SUPERVERTALER_HOST=memoq`, which tells it to look for memoQ’s connection instead of Trados’s. **Other MCP clients** (ChatGPT desktop, Claude Code, anything that runs a local STDIO server): unzip the server exe somewhere permanent and register it with that one environment variable set: ```json "supervertaler-memoq": { "command": "C:\\path\\to\\SupervertalerMcpServer.exe", "args": [], "env": { "SUPERVERTALER_HOST": "memoq" } } ``` The exe finds memoQ’s connection file in your Supervertaler data folder (`C:\Users\\Supervertaler\memoq\runtime\bridge.json`, or wherever you moved that folder). If you ever need to point it somewhere else, `SUPERVERTALER_BRIDGE_FILE` with a full path overrides it. ## The live document link memoQ’s MT plugin interface never shows a plugin the target text, the row you are on, or even the document’s name. memoQ’s **Preview SDK** – the interface its own PDF and video preview tools use – shows all three, live. So Supervertaler ships a small preview tool, `Supervertaler.MemoQ.Preview.exe`, which registers with memoQ exactly as the PDF preview does and forwards what memoQ sends it to the plugin. With it running, Claude sees your document as it actually is: every row’s current target, memoQ’s own row order, the document’s real name, and the row your cursor is on – and it can ask memoQ to **jump to a segment**. **Setting it up (once):** 1. Run `C:\Users\\Supervertaler\memoq\preview\Supervertaler.MemoQ.Preview.exe` – inside your Supervertaler data folder, where the plugin’s deploy puts it. A tray icon appears. 2. In memoQ, accept the **Preview tool connection request** for *Supervertaler*, leaving *Auto-start with memoQ* ticked. From then on memoQ starts the tool itself. 3. The tray icon reads *memoQ: connected · plugin: connected* once you click into a segment (that is what starts the plugin’s bridge). It appears under **Options → External preview tools** alongside any other preview tools; it can be disabled there like any of them. It draws nothing on screen – it is a link, not a preview. One thing to know: memoQ’s Preview SDK works in **paragraphs**, not segments. A paragraph that memoQ splits into three grid rows arrives as one unit with the whole paragraph’s source and target. The active-segment tool still reports the exact sentence your cursor is on, and jumps can target a sentence within a paragraph. Without the tool running, the tools below fall back to what the plugin captured from translation requests, and the two cursor tools say so rather than guessing. ## What it can and cannot do Everything the Trados server can do that memoQ *cannot* comes down to one fact: memoQ has no project API and no editor API for plugins. The live document link recovers the reading half of that; writing into the document still goes through you. The table is the honest map. | Tool | memoQ | Notes | | ---------------------------------------------------------------- | :---: | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `help` | ✓ | A menu of what you can ask, memoQ edition | | `get_project` | ✓ | Language pair, client/domain/subject, captured and live documents, what is staged | | `get_segments` | ✓ | With the live link: rows in memoQ’s order with source, **target**, and the active row marked. Without: the source segments captured from translation requests | | `get_active_segment` | ✓ | The row your cursor is on, with what is selected – needs the live link | | `go_to_segment` | ✓ | Asks memoQ to select a row – needs the live link | | `get_confirmed_pairs` | ✓ | Segments you have confirmed, via [Self-learning](/memoq/self-learning/) | | `lookup_term` / `add_term` | ✓ | Your Supervertaler [glossary](/memoq/terminology/), not memoQ’s term bases | | `stage_translations` | ✓ | **The write channel.** Translations wait until you Pre-translate | | `get_staged` / `clear_staged` | ✓ | Inspect and reset the staging area | | `list_prompts` / `get_prompt` / `save_prompt` | ✓ | The shared [prompt library](/memoq/prompt-editor/). A prompt Claude saves is recorded as drafted by the chat and marked for memoQ, so it does not appear in Trados’s list and the runtime treats its terminology the way it treats [AutoPrompt’s](/memoq/prompt-editor/#a-drafted-prompt-is-the-only-source-of-terminology) | | `list_supermemory_banks` | ✓ | Your [memory banks](#memory-banks), and how many articles each holds | | `get_supermemory_context` | ✓ | One bank’s brief, terminology and style, formatted for the model | | `search_supermemory` | ✓ | Full-text search inside a bank | | `update_segments`, `insert_into_active_segment` | ✗ | No write access to the editor – use `stage_translations` + Pre-translate | | `search_tm`, `search_studio_tm`, `compare_document_to_tm` | ✗ | memoQ’s TMs are not readable by plugins; confirmed pairs are the substitute | | `check_numbers`, `check_tags`, `check_nbsp`, `check_terminology` | ✓ | QA over the live document, paragraph by paragraph – needs the live link. `check_terminology` runs against the active Supervertaler glossary, so give it a project one: [Export glossary](/memoq/prompt-editor/#export-glossary-the-prompts-terms-as-the-project-glossary) from an AutoPrompt draft | | `find_inconsistencies` | ✓ | Repeated source paragraphs translated differently – needs the live link | | `run_verification` | ✗ | memoQ’s own QA cannot be run by a plugin; use memoQ’s Run QA | | `get_files`, `get_project_statistics`, `export_target` | ✗ | No project or file API | | `pretranslate` | ✗ | You press Pre-translate; that is the design | **Reading is complete with the live link; writing goes through you.** What you give up compared with Trados is Claude editing rows in place – and in memoQ the alternative, staging plus one Pre-translate, is a review step rather than a loss. ## Two channels for seeing the document Supervertaler captures segments in two ways, and it helps to know which is which when Claude reports what it can see: * **Translation requests** – every segment memoQ sends to Supervertaler as the MT engine. One Pre-translate captures the whole document, with its identity and metadata. This is the normal route. * **Terminology lookups** – every row your cursor lands on, through the [terminology plugin](/memoq/terminology/), *regardless of which MT engine is selected*. So a document you pre-translated with Google or from TM alone still becomes visible to Claude one visited row at a time. memoQ does not tell the terminology plugin which document a row belongs to, so these land in a per-language-pair bucket rather than under the document. `get_project` labels each captured document with its origin. ## Memory banks If you keep [SuperMemory](/trados/ai-assistant/super-memory/) banks – a folder per client, holding the brief, the terminology and the style rules you have settled on with them – Claude can read them here too. They live in one place for every Supervertaler product: ```plaintext C:\Users\\Supervertaler\memory-banks\ ``` Ask for what you want and Claude picks the tool: *“list my memory banks”*, *“read the Acme bank before you translate this”*, *“search the Acme bank for how we render ‘Vorrichtung’”*. Four things behave differently from the Trados plugin, and they are worth knowing before you rely on this: * **Claude is not told which bank is active.** You choose one in the [editor’s context bar](/memoq/prompt-editor/#what-memoq-is-using), and it is remembered per project – but that is what the plugin sends with its own translation requests. Over MCP the bank is named in the request instead, so say which one you mean. Claude will ask, or list them and let you choose. * **A name that does not exist is an error**, not a fall back to something else. Falling back would look exactly like success while feeding the model another client’s terminology, and nothing in the answer would say so. * **`_shared` is always underneath, and travels alone.** It is not a bank you select: whatever you do select is layered over it and wins wherever the two disagree – and when you select no client bank, `_shared` still goes on its own. Choosing “no client bank” is not the same as sending nothing. * **The answer is trimmed to about 6,000 tokens**, and whatever did not fit is listed under `trimmed` in the reply rather than dropped in silence. A tool result stays in the conversation and is re-sent on every following turn, so it is kept deliberately small – ask for a larger budget, or for one article by name, when you need the rest. Reading is all this does. Nothing writes into a bank from memoQ; you edit the files yourself, in Obsidian or any text editor. ## Troubleshooting **“Handshake file not found.”** memoQ has not created a Supervertaler engine yet in this session. Open a project and click into a segment with Supervertaler selected as the MT engine. **Claude says the project is empty.** Nothing has been captured yet. Run Pre-translate once, or visit some segments. **Staged translations do not appear after Pre-translate.** They are matched by exact source text. If you edited a source segment after Claude read it, the match fails – ask Claude to re-read and re-stage that segment. Also check that Supervertaler is the selected MT engine for the Pre-translate run. **`get_confirmed_pairs` is always empty.** Self-learning is not on. See [Self-learning translation](/memoq/self-learning/). # Prompt Library & Editor Supervertaler for memoQ translates with the instructions you give it. Those instructions can be typed straight into the settings dialog, or chosen from the **shared Supervertaler prompt library** – the same folder of prompts the Trados plugin uses, so a prompt tuned in one tool is available in the other. memoQ gives an add-in no window of its own, so the library cannot be a panel inside memoQ. Instead there is a small **Prompt Library editor**, opened from the settings dialog, that runs alongside memoQ. ## Choosing a prompt In **Resource console → MT settings → Supervertaler → Configure plugin**, the **Prompt** dropdown lists every translation prompt in the library, grouped by folder. Pick one and its text appears (read-only) in the **Instructions** box below. Choose **(instructions below)** instead to type your own; the box becomes editable. The dropdown stores *which* prompt you chose, not its text. Edit the prompt anywhere – in the editor, in the Trados plugin, in a text editor – and memoQ uses the new version on the next segment. Only prompts in the library’s **Translate** folder are offered. Proofreading and QuickLauncher prompts exist for other tasks and would produce commentary where a translation belongs. ## The editor Press **Edit…** beside the Prompt dropdown. * **Left:** the library as a tree, folders and prompts. Select one to open it. * **Right:** name, description, which product it is for, sort order, and the prompt text with Markdown headings and `{{PLACEHOLDERS}}` highlighted. * **Toolbar, left:** New, Save, Placeholder, AutoPrompt. * **Toolbar, right:** whether Pre-translate goes to Claude Desktop, the [Activity window](#the-activity-window), and Translation settings. Everything else is on the **File**, **memoQ**, **Settings** and **Help** menus. The **Claude Desktop** button on the right is a switch, not a command, and its caption says which mode is *on* rather than what pressing it would do. It decides whether Pre-translate spends your API key or hands the segments to the chat, so it is worth a glance before a long run. It is the same setting as **Settings → Pre-translate via Claude Desktop**, and as the checkbox in memoQ’s own dialog – change it anywhere and all three follow. Save with **Ctrl+S**. A prompt marked read-only in the library (the built-in defaults) can be read but not overwritten; make a copy under a new name instead. Any settings the file carries that the editor does not have a field for – tags, favourites, QuickLauncher flags set by the Trados plugin – are preserved untouched on save. The status line under the description says which ones the file has. ### What memoQ is using The bar under the toolbar opens with the **memoQ project** these apply to, because everything on it is recorded against a project. When it reads **no project yet**, in red, memoQ has not sent a translation request and the plugin does not know where it is – usually because Supervertaler is not selected as the MT engine in a newly created project. A memory bank chosen at that moment is filed against whichever project came before, silently, so it is worth a glance before changing anything. Then three things memoQ will apply to every translation: **Prompt**, **Glossary** and **Memory bank**. Click any of them to change it. Each opens a list with a filter box rather than a dropdown menu, because all three grow with the work – a prompt library reaches forty entries quickly, and a bank per client does the same. All three can also be set to nothing: the glossary list has a **(none)** row, and a **Browse…** row at the end for a glossary that lives outside the glossaries folder. These are the choices that change between jobs, which is why they are here rather than in Translation settings: they are what the model knows before it is shown a segment. Each is also the same setting memoQ’s own dialog shows, so either place can change it. The memory bank is remembered **per project**. Choose one while working on a job and it comes back when you return to that job – and a project you have never chosen one for does *not* inherit the last one, because a bank carries one client’s terminology and the wrong one is worse than none. What such a project gets instead is the row called **(no client bank – shared defaults only)**. Your `_shared` bank still travels: it is where the material that applies to every job regardless of client lives, so it is never switched off by not choosing a client. See [Memory banks](/memoq/mcp-server/#memory-banks) for where they live. A bank is sent whole with **every** translation request, up to about 32,000 tokens, and to AutoPrompt up to 40,000. Anything that does not fit is dropped by priority and named in the [Activity window](#the-activity-window) rather than lost quietly – if you see a file listed there, that is the budget, not a fault. The cost of carrying it is small because the same text is sent every time and providers cache it: on one 370-segment run the bank and prompt together came to 45,870 tokens, and caching turned roughly $10 into roughly $4. ### Placeholders Prompts use `{{SOURCE_LANGUAGE}}` and `{{TARGET_LANGUAGE}}` rather than naming languages, so one prompt serves every language pair. memoQ fills them in per project. **Insert placeholder** lists the ones memoQ can fill. A placeholder memoQ cannot fill – `{{SOURCE_SEGMENT}}`, say, which only the Trados plugin provides – is shown in red, and the editor warns that it will reach the model as empty text. Do not use those in a prompt meant for memoQ. ### Which product a prompt is for **Available in** can be *both*, *trados* or *memoq*. memoQ’s dropdown hides prompts marked for Trados only. Leave it on *both* unless a prompt genuinely depends on something one product cannot supply. A prompt tied to one product says so in its **filename**: `Patent claims EN-NL [memoQ].md`, `Define [Trados].md`. Prompts available to both carry no marker, so the absence of one is itself readable – which is the point, because in Explorer the metadata header is not visible and every prompt otherwise looks alike. The marker is written from the **Available in** field on every save and stripped again on every read, so it is a label rather than a setting. Renaming the file in Explorer does not change which product a prompt is for, and the next save puts the old marker back: change the field, not the filename. The editor’s tree and the Prompt dropdown show the same thing in words. ## Where the library lives `C:\Users\\Supervertaler\prompt_library\` – one Markdown file per prompt, with a small metadata header. **Open folder** in the editor takes you there. The files are plain text; nothing stops you editing them directly, and a folder synced between machines carries the whole library with it. ## AutoPrompt: drafting a prompt for the open project Press **AutoPrompt…** in the editor’s toolbar, or choose it from the **memoQ** menu. Supervertaler reads the document you are translating, your glossary hits in it and anything you have already confirmed, and has the AI write a prompt tailored to that job – domain, register, a locked glossary, the lot. The result is saved under **Translate** and opened for you to review; then pick it from memoQ’s **Prompt** dropdown. Before it runs you choose the document (if several are captured), and can add a briefing – client, audience, style, what to avoid – which the AI treats as authoritative. The briefing is the one input nothing else supplies: the filing route, a discrepancy you already know about, anything true of this job that is not in the document or the memory bank. The two checkboxes grey out when they have nothing to offer – no glossary is active, or nothing has been confirmed in this document yet – and say which it is, rather than sitting there ticked and doing nothing. **Preview context…** shows you exactly what will be sent, before anything is sent: the extract from your document, the glossary hits, the segments you have confirmed, the briefing you typed, and the instructions the AI is given about writing a prompt for memoQ. It makes no API call and costs nothing, and the briefing box stays open behind it – so the loop is look, add what is missing, look again, then generate. Three things to know: * **memoQ must be running with a Supervertaler engine active**, and the document must have been captured – one Pre-translate does it (free, with the [Claude Desktop box](/memoq/mcp-server/#the-checkbox) ticked). The plugin only sees what memoQ has sent it. * **It uses the provider, model and API key from your Supervertaler settings.** Two calls: a short one to classify the document, then a long one to write the prompt. Expect a minute or two. * **The prompt is written for memoQ, not copied from the Trados recipe.** Single-segment lookups are handled as well as batches; tag markers must be reproduced exactly; translations you have confirmed outrank the prompt’s own glossary; and it is kept to 1,500–3,000 words because memoQ re-sends the whole prompt with every ten-segment request. Draft it again later in the job and it gets better: by then it can see what you have confirmed, which is stronger evidence of how you want *this* document translated than the source text alone. ### A drafted prompt is the only source of terminology A prompt AutoPrompt wrote ends in a locked-terms table chosen for this document. So while one is selected, **the glossary’s preferred renderings are not sent to the model as well** – two lists of terminology that were never written to agree, with nothing saying which wins, is a worse position than one list. **Forbidden terms still go.** A preferred rendering is advice, and two sources of advice can contradict each other confusingly; “never use this word” is a constraint, and there are few of them. So they travel whatever prompt is selected. Nothing else about the glossary changes: it still drives the terminology pane, the QA check and AutoPrompt’s own reading of the document. Only the per-request injection stops, and the [Activity window](#the-activity-window) says so once per prompt. The consequence worth remembering: a term you forbid **after** a prompt was drafted is enforced immediately, but a preferred rendering you add afterwards is not – draft the prompt again to take it in. **Export glossary** is the other half of that loop: derive the glossary *from* the prompt and the two cannot contradict each other in the first place. Prompts saved from the chat over [MCP](/memoq/mcp-server/) count as drafted too, and are marked for the product you were connected to. ### List numbering reaches the model as structure The letters on the steps of a claim – a), b), c) – are not text. Word generates them from the paragraph’s list settings, so memoQ’s grid does not contain them and neither does anything the plugin is sent. Shown six unlabelled sentences followed by *steps a. to f.*, a model will flag the reference as a possible source defect, and it will do so on every lettered list in every document. With the [live document link](/memoq/mcp-server/#the-live-document-link) connected, the plugin knows which file memoQ imported, reads the numbering out of it – counted over the whole document, exactly as Word renders it, restarts and all – and sends each paragraph’s marker in front of its first segment as `[#e)]`. The prompt tells the model this is structure: use it to resolve cross-references and keep list items parallel, never translate it, never reproduce it. Every reply is checked before it reaches the document, and an echoed marker is removed, with a line in the [Activity window](#the-activity-window) saying so – that line is your evidence, per model, that the rule is being obeyed. There is no switch for it. Without the live link, or for a document with no lists, the model is instead told that numbering is supplied by the document and not to flag its absence. The Activity window says which of the two happened, once per document. On a project checked out from a server the file memoQ names is on the project manager’s machine, not yours; locate it once in [FigureLens](#where-the-documents-come-from) and the numbering is read from your copy. ### Translator comments Where a note is genuinely necessary – an ambiguity in the source, a term that could go two ways, a probable defect in the original – a drafted prompt has the AI put it inline at the end of the target as a `[[TC: …]]` marker. Supervertaler for Trados uses the same form, so a prompt written for one product reads correctly in the other. Nothing extracts these for you, and that is deliberate. You read them in the grid as you review, decide which are worth keeping, turn those into real memoQ comments on the segment, and delete the marker from the text. Search for `[[TC:` to find them all. ## FigureLens: what the figures show The model sees a document’s text and not its pictures. A claim that names *part 12* is translated by a model that has never seen part 12, and a figure’s caption is often the only description of it anywhere in the text. **FigureLens…** on the toolbar (also **memoQ → FigureLens…**) is the panel that closes that gap, in two steps. It is named for what it does beside [TermLens](/memoq/terminology/): that one shows the model the terms in a segment, this one shows it the pictures. **Step 1 – Extract images** copies every image out of the documents into the active memory bank’s `figures\` folder, named after their figure numbers – `Figure 01.png`, `Figure 02.png` – so a folder of drawings reads like the document. Free, no AI. The panel says how the labels were arrived at: paired by position and checked, taken from nearby text, or withheld when it could not tell. **Step 2 – Describe images with AI** shows each image to the model, together with what the text says about it, and saves the descriptions as `figures.md` in the memory bank – one paid request per image, and the panel states the count and the provider before you click. Every prompt reads that file from then on, so read it first: a wrong caption would be invisible and everywhere. Reference signs the model reads in a drawing that appear nowhere in the text are listed at the end, because that is a defect worth raising with the client before filing. **Describe from the text only** is the free alternative – what the document itself says about each figure, without looking at the images – and either replaces the other, after asking. The images folder is not chosen: it is inside the memory bank, because that is where `figures.md` goes and the two belong together. With the shared bank or no bank active, the Result line offers to create a bank named after the memoQ project and switch to it. ### Where the documents come from memoQ never keeps the original file in its project folder – a local project stores the filename as an empty placeholder, a project checked out from a server stores only memoQ’s own data – so the panel works from the file memoQ imported, wherever that was: * On a **local project**, the [live document link](/memoq/mcp-server/#the-live-document-link) reports the path memoQ imported each document from, and the panel finds the file there without any setting. * On a **server project**, memoQ records the path the file had on the project manager’s machine, which does not exist on yours. The panel lists those documents as *not on this computer* and offers **Locate the original document…** – point it at the copy you were sent, once, and it is remembered for that document in `C:\Users\\AppData\Local\Supervertaler.memoQ\document-files.txt`. Locating a document here also switches on [list numbering](#list-numbering-reaches-the-model-as-structure) for it. * **Add a document file…** is for a Word file memoQ has said nothing about at all. Its images are read from the file directly; the file is remembered for the active memory bank. Supervertaler finds the project folder by asking memoQ where it keeps its projects – the custom folder set under **Options → Locations → Projects**, the default `C:\Users\\Documents\My memoQ projects` when you have not set one, and memoQ’s own register of every project, which names each one’s actual folder. Moving your projects folder therefore needs nothing here, and projects left behind in the old location are still found. **Document images report** at the bottom writes a Markdown listing of every image in every document – label, size, caption, the text around it – into the memory bank and opens it. No AI call. ## Export glossary: the prompt’s terms as the project glossary An AutoPrompt draft ends with a locked-terms table – a dozen or so renderings chosen for this document. That table is exactly what the [terminology plugin](/memoq/terminology/) and the `check_terminology` QA tool should work from: a general glossary flags *application → aanvrage* in every paragraph of a software patent, a project glossary knows better. Choose **memoQ → Export this prompt’s terms as a glossary** with the prompt open. Supervertaler reads every table in it that names a source and a target column, turns notes of the form *never “apparatus”* into forbidden entries, and writes a tab-separated glossary file to `C:\Users\\Supervertaler\memoq\glossaries\.txt`. Answer yes when it asks and that file becomes the active glossary immediately, whether or not memoQ is running. The file is plain text – edit it freely; the plugin re-reads it whenever it changes. Any prompt with a table laid out the same way works, not only AutoPrompt’s. ### Settings **Settings → Translation settings** holds how Supervertaler translates: provider, model, endpoint, parallel requests, segments per request, and whether termbase hits and surrounding segments are sent to the model. These are the same settings as memoQ’s own Supervertaler dialog, reading and writing the same file, so either place can change them and both show the same values. The **Model** list is short on purpose: three to five models per provider, each with a line saying what it is for. A provider’s own catalogue runs to thirty or forty entries – image models, speech models, dated snapshots of the same model – and a list like that is one nobody in a hurry can choose from. A model that has been superseded is removed rather than annotated, so what is left is what is worth using today. **Fetch list** asks the provider for its full list, using the API key below. **Show all models** then shows everything it returned under the short list, which is where to look for a model released after your copy of Supervertaler was built. The fetched list is remembered, and so is the tick, so this is a decision you make once. The line under the button says when the list was last fetched and how much of it is beyond the short list. The box stays typeable throughout, so a gateway, a private deployment or a model that appears in neither list can be entered by hand – and a model already saved in your settings keeps working whether or not it is in the list on screen. Changing the provider changes three things together: the model list, the model itself – to that provider’s first recommendation, since a model belonging to another provider can only fail – and the API key, which is re-read for the provider you have just chosen. A key you type here is remembered per provider while the window is open. memoQ’s own **Configure plugin** dialog has the same three controls, reading and writing the same settings, so it does not matter which one you use. **Segments per request** can only lower what memoQ does, not raise it – memoQ hands a plugin about ten segments at a time during Pre-translate, however high this is set. Lowering it is still worth doing if a model keeps returning fewer translations than it was sent. **Settings → Pre-translate via Claude Desktop (MCP)** is on the menu itself, and on the right of the toolbar, because it is the one that gets flipped between jobs rather than set once. See [MCP server](/memoq/mcp-server/). ### API keys live in one file Every Supervertaler product reads the same file: ```plaintext C:\Users\\Supervertaler\settings\api-keys.json ``` One key per provider, plain text, editable in Notepad. A key pasted here works in Supervertaler for Trados and Supervertaler Sidekick as well, and rotating one means changing one line in one place. Before this file existed there were three dialogs in three products each keeping their own, which is how an hour goes missing to a key for one service pasted into another’s box. Keys are stored under the provider ids `claude`, `openai` and `gemini`. Sidekick keeps its machine-translation keys in the same file under their own names – note that `google` there is Google Translate, not Gemini. Plain text is deliberate, and the same choice Supervertaler for Trados has always made: a key that can be rotated by pasting a line into a text file is a key that actually gets rotated, and anyone who can read that file can already read everything else in your profile. The **API key** box shows the key for the provider you have selected and writes back to that file. If you had a key configured before the file existed – in memoQ’s own settings, or in Trados’s – it is copied in the first time Supervertaler needs it, so there is nothing to do. If you paste a key that plainly belongs to another service, the line under the box says so as you type: *This is an OpenAI key, not an Anthropic one.* That is worth more than the provider’s own answer, which is that the key is incorrect. ## The Activity window memoQ’s Pre-translate dialog is modal and says only *Processing*, for as long as the run takes: no engine, no model, no count, and no sign when something is wrong. **memoQ → Activity…**, or **Ctrl+L**, opens a window that shows what Supervertaler is actually doing. It is a window of its own rather than a panel so that it can sit over memoQ while that dialog holds the screen. Tick **Keep on top** and you can watch a Pre-translate run from the first batch to the last. What it shows: the engine and model each project starts with, the glossary as it loads and how many terms came out of it, warnings when the selected prompt or glossary faces the opposite language pair, every batch with the segments sent, the segments returned and the glossary terms matched, AutoPrompt drafts, and anything that failed. A batch that comes back short is called out rather than logged flatly, because that is the failure that quietly shifts every translation after it. Three lines are worth knowing by sight: * **Bank** – which memory bank a project switched to, and once per job how much of it is being sent. If it ends with a file listed as *not sent*, that is the budget trimming by priority, not a fault. * **Terminology** – said once when a drafted prompt is holding the glossary back, so a quiet change to what reaches the model is never silent. * **The token count on each batch** – `tokens: in 1,041 (cache write 45,870) out 1,233`. The prompt and the bank are identical on every request of a run, so providers cache them: the first batch writes, the rest read at a tenth of the rate. If *cached* never appears across a long run, something is re-sending the block at full price. **Show everything** un-hides the per-request diagnostics – memoQ’s capability probes, lookup sessions, single-segment translations – which are what you want when something is wrong and noise the rest of the time. The window reads the plugin’s own log, `C:\Users\\AppData\Local\Supervertaler.memoQ\plugin.log`, rather than being fed by the plugin. So it shows what happened before you opened it, it works whether or not memoQ is running, and closing it costs nothing. Its position and size are remembered. ## Drafting prompts with Claude If you use the [MCP server](/memoq/mcp-server/), Claude can write into this library: *“Draft a translation prompt for this project and save it.”* It reads the captured document, your confirmed segments and your glossary, saves the result as a new prompt, and you pick it from the dropdown. The editor is where you review and tune what it wrote. # Self-learning Translation Supervertaler remembers the segments you confirm, and shows the most relevant ones to the model when it translates later segments in the same document. This is what makes the second half of a document read like the first. Settle on a rendering once, confirm it, and everything that follows is translated with your choice in front of the model rather than against a blank slate. ### Turning it on **Resource console → MT settings → your resource → Settings tab → Self-learning MT → Supervertaler.** Then restart memoQ. Until this is set, memoQ never passes confirmed segments to the plugin, and Supervertaler translates without memory. ### What gets remembered Only segments **you confirm**. Not the AI’s raw output. That distinction is deliberate. Feeding a machine its own guesses compounds its mistakes; the value of these examples is precisely that a human approved them. For each confirmation Supervertaler stores the source text and the target text, and nothing else – no tags, no formatting, no metadata. Re-confirming a segment replaces the earlier version, so your latest decision is the one that counts. ### How much is sent to the AI Not all of it. For each new segment, Supervertaler picks the **five** stored pairs sharing the most vocabulary with it. A pair with no words in common is never sent. So a document with hundreds of confirmed segments still contributes only a few short examples per request – the ones most likely to matter. ### What is stored, and where Memory is kept per document *and* language pair, and written to: ```plaintext C:\Users\\AppData\Local\Supervertaler.memoQ\document-memory\ ``` It survives closing memoQ, so picking a job back up the next morning keeps yesterday’s decisions. Limits: 500 pairs per document, 200 documents, and anything untouched for 60 days is discarded. This is client text on your disk These files contain source and target segments from real work. They are stored under your own user profile, are not synced anywhere, and never leave your computer. The plugin’s options dialog shows how many documents and how much space are held, with a **Forget stored context** button that deletes all of it. Your translations in memoQ are not affected. ### What it is not This is not adaptive machine translation in the sense that ModernMT or Lara mean it. No model is trained or fine-tuned, and nothing is sent to Supervertaler – the examples are simply included in the request to the AI provider you chose, alongside the segment. The practical differences: it works within a document rather than across your whole history, it is lost if you clear it, and it is entirely inspectable. Nothing happens that you cannot see in the prompt. # Terminology Supervertaler reads a glossary file and uses it in two places at once: memoQ’s terminology pane, and the AI’s prompt. ### Why a file, and not a memoQ term base memoQ does not let a plugin read its own term bases. A term base attached to your project is visible to you and to memoQ’s QA, but not to a machine-translation plugin – so a term marked forbidden in memoQ will not, on its own, stop the AI using it. Supervertaler works around this by being a terminology source in its own right. Terms in its glossary reach the model because Supervertaler puts them there. ### Setting it up **Options → Terminology plugins.** 1. Tick **Perform terminology plugin lookups while working in the translation grid**. Nothing happens until this is on. 2. Find **Supervertaler terms** in the list. It reads *Not configured* until a glossary is set. 3. Click its **Options**, choose your glossary file, press OK. 4. Tick **Enable plugin**. Restart memoQ. The same glossary setting is reachable from the translation engine’s options dialog – both halves of the plugin read one file. ### Which glossary is active One setting, three consumers: the terminology pane, the prompt sent to the model, and the terminology QA check all read the same file. It is shown in four places, so you never have to guess which one is answering: * **The engine’s options dialog** (Options → Machine translation → Supervertaler → Options) has a *Glossary* row naming the active file with its full path, in red if the file has gone missing. *Change…* opens the same chooser the terminology plugin uses. * **Every hit in Translation results** carries the glossary’s file name in its grey footer (*Supervertaler · patent eng-dut.txt*). * **The [prompt editor](/memoq/prompt-editor/)** names it in its status bar, where clicking it changes it, and works with memoQ closed. * **Claude’s project report** (`get_project` over the [MCP server](/memoq/mcp-server/)) includes the path under *activeGlossary*. Exporting a glossary from the [prompt editor](/memoq/prompt-editor/) makes that file the active one immediately; the options dialog and the footer show the new name on the next lookup. ### What you get **In the grid.** Matched terms are highlighted in the source segment: green for approved terms, red for forbidden ones. **In Translation results.** Each match appears as an entry showing the target term, the source term it matched, and – for a forbidden term – the wording struck through under *Do not use*. **In the prompt.** Approved terms are sent as the client’s preferred wording; forbidden terms as absolute constraints. **For Claude, if you use the [MCP server](/memoq/mcp-server/).** Every row your cursor lands on is looked up by this plugin whatever MT engine is selected, and Supervertaler remembers each one – so a document you pre-translated with Google or from TM alone still becomes visible to Claude as you walk through it. Claude can also read and add glossary entries directly (*“we agreed* draagarm *=* support arm *– add it”*). ### Preferred, not mandatory Approved terms are given to the model as a strong steer it may override when an entry is clearly wrong for the sentence at hand. That asymmetry is deliberate, and it comes from a real failure. A patent term base may quite correctly render *applications* as *aanvragen* – in the sense of a patent application. Told to use terminology verbatim, the model translated “Mashup applications” as “Mashup-aanvragen”, which is nonsense. A translator treats a term base as guidance they may set aside with reason, and the model is asked to do the same. **Forbidden terms are not softened.** They are stated as absolute, because that is what a forbidden term is for. ### Format See [Glossary format](/memoq/glossary-format/). # Troubleshooting ### Supervertaler does not appear at all Check `Supervertaler.MemoQ.dll` is in memoQ’s `Addins` folder, and that you answered **Yes** to the unsigned-plugin prompt on startup – its default button is *No*. If memoQ was recently upgraded to a new major version, the add-ins need copying into the new program folder. See [Installation](/memoq/installation/). ### It translates, but never learns **Self-learning MT** is not set. Resource console → MT settings → your resource → **Settings** tab → **Self-learning MT** → *Supervertaler*, then restart memoQ. Advertising the capability only makes the engine eligible; memoQ does not send confirmations until it is actually selected there. ### Terminology is not showing Three things must all be true, under **Options → Terminology plugins**: 1. **Perform terminology plugin lookups while working in the translation grid** is ticked 2. **Supervertaler terms** does not read *Not configured* – i.e. a glossary file is set 3. **Enable plugin** is ticked for it ### The glossary loads no terms Almost always spaces where tabs should be. The glossary options dialog reports how many terms it parsed; if that reads zero with a file selected, open the file in an editor with whitespace visible and check the separators. ### The panel names the wrong project, or nothing reaches the model Supervertaler learns which project is open from two channels: a translation request from memoQ, and the [live document link](/memoq/mcp-server/#the-live-document-link), which reports every document memoQ shows. With the link connected, the panel follows within a couple of seconds of your clicking into a segment of the new project. Without it, the plugin has only translation requests – an add-in cannot ask – so until the first request of a session the panel still names the last project memoQ *did* ask about, and a memory bank chosen at that moment is recorded against that earlier project. **Click the project name** in the panel (or **memoQ → Sync with memoQ now**) to take the project from the document memoQ is showing this instant. If it cannot, it says why: the live link is not connected, memoQ has not reported a document yet, or no project folder holds that document. Where Supervertaler looks is memoQ’s own answer – the custom folder set under **Options → Locations → Projects**, the default `C:\Users\\Documents\My memoQ projects` when none is set, and memoQ’s register of every project, which names each one’s actual folder – so moving your projects folder needs no setting here, and projects left behind in the old location are still found. If clicking into a segment does not update the panel, memoQ is not calling the engine at all. Open **Project home → Settings → MT settings** and look for the line **“MT plugins are currently disabled.”** A project checked out from a memoQ server can have MT plugins switched off by the project manager, and there is no client-side setting that overrides it. The terminology provider still works in such a project, because it is not an MT plugin; translation through Supervertaler does not, in any mode, and neither does staging from Claude Desktop, which enters the grid through the same engine. Ask the project manager to allow MT plugins, or take the document out through a bilingual export. If you chose a memory bank while the panel was stale, open the bank chooser again once the right project is shown – that records the choice against the right project – and check the previous project’s row in `C:\Users\\AppData\Local\Supervertaler.memoQ\memory-bank-projects.txt`, one project GUID per line, in case it now names a bank it should not. ### The log The **Activity** window in the prompt editor (**memoQ → Activity**, or Ctrl+L) shows the same log live. On disk it is: ```plaintext C:\Users\\AppData\Local\Supervertaler.memoQ\plugin.log ``` with a fallback at `C:\Users\\AppData\Local\Temp\Supervertaler-memoQ.log` if that folder cannot be written. It records what memoQ asked for and what was sent – segment sizes, how many glossary terms matched, how many remembered segments were used, and any errors. It does not contain the text of your translations. A typical healthy line: ```plaintext translate: 199 src chars, 0 tag(s) -> 239 target chars, 0 tag(s) | recall: used 2 of 7 held | terms: 7 ``` meaning: a 199-character segment; two remembered segments and seven glossary terms sent with it; a 239-character translation returned. # Supervertaler for Trados Supervertaler for Trados is a plugin for **Trados Studio 2024+** that brings Supervertaler’s terminology and AI features directly into the Trados editor. It runs natively inside Trados Studio as a set of dockable panels, so you never have to leave the editor. ![]() ### Getting Started Screencast New to Supervertaler? Watch the [Getting Started screencast](https://www.youtube.com/watch?v=bOIwMAoP7xc) (16 min) for a walkthrough of all the basics – TermLens, prompt generation, AI translation, the Chat window, and more. [Supervertaler for Trados – Getting Started (16 min)](https://www.youtube.com/embed/bOIwMAoP7xc) ### Key Features #### TermLens (Inline Terminology) Live terminology display that shows the source text word by word, with termbase translations underneath each matched term. Colour-coded by termbase type: * **Blue** for regular termbase matches * **Pink** for project termbase matches (higher priority) * **Yellow** for non-translatable terms * **Green** for MultiTerm termbase matches (`.sdltb` files attached to your Trados project, or `.ttb` termbases in Trados Studio 2026) Numbered badges let you insert terms with **Alt+1** through **Alt+9**. Project termbases are detected automatically from your Trados project and are read-only. On Trados Studio 2026 these are the new `.ttb` termbases –see [Trados Studio 2026 & .ttb](/trados/studio-2026/). #### Supervertaler A conversational AI chat panel that is aware of your current segment, matched terminology, and TM matches. Ask questions about translation choices, get alternative phrasings, or request explanations –all without leaving Trados. #### Batch Translate Translate multiple segments at once using AI. Choose a scope (empty segments, all segments, filtered segments), pick a prompt, and let the AI work through your file. Progress is shown in real time. #### Prompt Library A built-in Default Translation Prompt and Default Proofreading Prompt to get you started, plus QuickLauncher prompts for common tasks. Create your own custom prompts in the Prompt Manager – duplicate the default and tailor it to your domain. #### Termbase Management Create, edit, and import termbases in Supervertaler’s `.db` format. Quick-add terms with keyboard shortcuts, mark terms as non-translatable, and manage multiple termbases per project. #### SuperMemory The client decisions you cannot look up: which term they insist on, what they rejected last time, how they want things phrased. Each memory bank is three Markdown files you write and edit yourself - a brief, a terminology table, and style rules - plus a shared bank of house defaults that every client bank can override. The AI reads the active bank on every translation. Keep one bank per client or domain and switch from the toolbar dropdown; add a term while translating with Ctrl+Alt+M. [Learn more →](/trados/ai-assistant/super-memory/) ### System Requirements | Requirement | Version | | -------------- | ------------------- | | Trados Studio | 2024 (v18) or later | | Windows | 10 or 11 | | .NET Framework | 4.8 | There are two builds: one for **Trados Studio 2024** (MultiTerm `.sdltb` termbases) and one for **Trados Studio 2026** (`.ttb` termbases). Install the build that matches your Studio version –see [Trados Studio 2026 & .ttb](/trados/studio-2026/). ### Shared Termbase Format Supervertaler for Trados uses the same SQLite-based termbase format (`.db`) as [Supervertaler Workbench](https://docs.supervertaler.com/workbench/). Termbases created in either tool are fully compatible – you can open the same `.db` file in both applications. ### Context-Sensitive Help Press **F1** at any time to open context-sensitive help for the panel or dialogue that currently has focus. ### Next Steps | | | | ---------------------- | ------------------------------------------------------------- | | **Installation** | [Install the plugin →](installation.md) | | **Getting Started** | [Set up termbases and AI →](getting-started.md) | | **TermLens** | [Inline terminology display →](termlens.md) | | **Supervertaler** | [Chat with AI in Trados →](ai-assistant.md) | | **MultiTerm Support** | [Use MultiTerm termbases in TermLens →](multiterm-support.md) | | **Batch Translate** | [Translate segments in bulk →](batch-translate.md) | | **Keyboard Shortcuts** | [All shortcuts at a glance →](keyboard-shortcuts.md) | # Supervertaler Assistant The Supervertaler Assistant is a conversational chat panel that runs inside Trados Studio as a separate dockable panel. It is context-aware: it automatically includes your current source and target text, matched terminology, and TM matches in every request, so the AI can give you informed answers about the segment you are working on.  ## Opening the Panel The Supervertaler Assistant lives in its own dockable panel. To open it, go to **View > Supervertaler Assistant**. You can dock the panel on the right side, bottom, or as a floating window. Trados remembers the panel position between sessions. ## Chat Type a message in the input field at the bottom and press **Enter** to send. The AI will consider your current source text, target text, matched terminology from your termbases, and TM fuzzy matches when responding. | Action | How | | --------------------------- | ------------------------- | | Send a message | Press **Enter** | | Insert a line break | Press **Shift+Enter** | | Stop a response in progress | Click the **Stop** button | ### What You Can Ask Because the assistant has access to your current segment context, you can ask things like: * “Translate this segment” * “What is the difference between these two translations?” * “Is this terminology correct in a legal context?” * “Suggest a more formal alternative” * “Explain this source text” ### Chat History The conversation is saved automatically after every message and restored the next time Trados starts. Your history persists until you explicitly clear it. To clear the history, click the **Clear** button in the chat toolbar. Note Chat history is stored in `~/Supervertaler/trados/chat_history.json`. It is a single global history – not per project or per file. ### Right-Click Menu Right-click any assistant response bubble to access: | Action | Description | | ----------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Copy** | Copies the raw Markdown to the clipboard, preserving tables and formatting | | **Apply to target** | Inserts the plain text (Markdown stripped) into the active target segment | | **Save as Prompt…** | Saves the response as a reusable prompt template | | **Save to memory bank** | Saves the question + response into the active bank’s `reference/` folder so a useful answer is not lost. Nothing reads it automatically - fold anything worth keeping into `brief.md`, `terminology.md` or `style.md`. | If you select text within a bubble before right-clicking, **Copy** and **Apply to target** operate on the selection only. ## Features | Feature | Description | | ---------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ | | [**Context layers**](/trados/context-layers/) | The ten layers of context sent with every request – project, segment, TM, terminology, SuperMemory, list numbering, figures | | [**File Attachments**](/trados/ai-assistant/file-attachments/) | Attach images and documents (PDF, DOCX, XLSX, TMX, etc.) for additional context | | [**Studio Tools**](/trados/ai-assistant/studio-tools/) | Query your Trados Studio projects, TMs, termbases, and statistics using natural language | | [**Incognito Mode**](/trados/ai-assistant/incognito-mode/) | Anonymise project names, file paths, and personal data in AI responses for safe sharing | | [**Providers and Models**](/trados/ai-assistant/providers/) | Supports 7 AI providers including OpenAI, Claude, Gemini, Grok, Mistral, Ollama, and custom endpoints | | [**Supervertaler Bridge**](/trados/ai-assistant/supervertaler-bridge/) | Localhost-only HTTP service behind the [MCP Server](/trados/mcp-server/): how Claude Desktop and other AI apps read and write your live Trados project | ## See Also * [QuickLauncher](/trados/quicklauncher/) – One-click prompt shortcuts * [Batch Translate](/trados/batch-translate/) – Translate multiple segments at once * [AI Settings](/trados/settings/ai-settings/) – API keys, model selection, context options * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) # Chat The **Chat** tab is the conversational heart of the Supervertaler Assistant – a context-aware chat that already knows your current segment, its terminology and TM matches, so you can ask questions and get answers grounded in the document you are translating. (See the [Overview](/trados/ai-assistant/) for opening the pane and the basics of chatting.) Two features extend what a chat can do: * **[File Attachments](/trados/ai-assistant/file-attachments/)** – attach images or documents (a reference PDF, a style guide, a screenshot of the source layout) to a message for the AI to use as context. * **[Studio Tools](/trados/ai-assistant/studio-tools/)** – ask the assistant about your Trados installation in plain language (projects, translation memories, termbases, templates) and it looks the answer up for you. # File Attachments The Supervertaler Assistant supports attaching both images and documents to your messages. Use the **paperclip button** next to the chat input, or drag and drop files directly onto the chat area. ## Images Attach images for visual context – for example, a screenshot of the source document layout, a reference image, or a table that is hard to describe in text. Images are sent to the AI using each provider’s native vision API. | Method | How | | ------------- | --------------------------------------------------- | | Paste | Press **Ctrl+V** with an image on the clipboard | | Drag and drop | Drag an image file into the chat input area | | Browse | Click the paperclip button and select an image file | Supported image formats: PNG, JPEG, GIF, WebP, BMP. Up to **5 images** per message, **10 MB** maximum per image. ## Documents Attach documents to provide the AI with additional reference material – for example, a client style guide, a termbase in spreadsheet form, a reference PDF, or a translation memory export. The text content is automatically extracted from the document and included in your message as context. | Method | How | | ------------- | ----------------------------------------------------- | | Drag and drop | Drag a document file into the chat input area | | Browse | Click the paperclip button and select a document file | The chat bubble shows a compact summary (file name and size) instead of the full extracted text, keeping the conversation readable. ### Supported Document Formats | Category | Formats | | ----------------- | ------------------------------ | | Documents | DOCX, DOC, PDF, RTF | | Presentations | PPTX, PPT | | Spreadsheets | XLSX, XLS, CSV, TSV | | Translation files | TMX, SDLXLIFF, XLIFF/XLF, TBX | | Text and markup | TXT, Markdown, HTML, JSON, XML | Note Up to **5 documents** per message, **20 MB** maximum per file. Very large documents are automatically truncated to avoid exceeding AI context limits. Legacy binary formats (DOC, XLS, PPT) use best-effort text extraction – for best results, save as the modern format (DOCX, XLSX, PPTX) first. Tip **Tip:** Attaching a client style guide or reference document alongside your translation question gives the AI much better context for providing accurate, style-consistent suggestions. ## See Also * [Supervertaler](/trados/ai-assistant/) – Overview * [Context layers](/trados/context-layers/) – What context is sent automatically # How the assistant works Whatever you do in the Supervertaler Assistant – chat, a batch translation, a proofread, or an AutoPrompt – it works the same way under the hood: it gathers **context** from your project and sends it to an **AI provider**, then returns the result. These pages explain what the AI sees and how to control it: * **[Context layers](/trados/context-layers/)** – exactly what Supervertaler puts in front of the AI, in ten layers (current segment, surrounding segments, terminology, TM matches, SuperMemory, the document’s list numbering, descriptions of its figures) and how to tune how much is included. * **[Providers and Models](/trados/ai-assistant/providers/)** – which AI you talk to (Claude, OpenAI, Gemini, local models …), and how to set up your API key and choose a model. * **[Incognito Mode](/trados/ai-assistant/incognito-mode/)** – a privacy filter that anonymises project names, file paths, TM names and other identifying data in the AI’s responses, so you can screen-share, record or post screenshots without exposing client information. # Incognito Mode Incognito Mode tells the AI to **anonymise all personal and project data** in its responses. When enabled, project names, file paths, TM names, user names, and other identifying information are automatically replaced with plausible placeholders – so you can share your screen, record videos, or post screenshots without worrying about exposing confidential client data. 🕵️ Think of it as a privacy filter for your AI chat. ## When to Use It | Scenario | Example | | ------------------------ | ----------------------------------------------------------------------------- | | **Screen sharing** | Presenting to colleagues or in a webinar while working on a real project | | **Recording demos** | Making tutorial videos that show real workflows without real client names | | **Forum posts** | Sharing a helpful AI response in a community without revealing client details | | **Client presentations** | Showing how the tool works without exposing other clients’ data | | **Training** | Onboarding new team members on live projects | ## How It Works When Incognito Mode is enabled, the AI receives an instruction to replace all identifying data with anonymised equivalents. For example: | Real data | Anonymised | | ---------------------------- | -------------------------------------- | | Acme Corporation | Client Alpha | | D:\Jobs\ACME\Q1\_report.docx | D:\Projects\Client Alpha\document.docx | | Jane Smith | User A | | ACME\_NL-EN.sdltm | Client\_Alpha\_NL-EN.sdltm | The AI uses **consistent replacements** within a conversation, so if “Acme Corporation” becomes “Client Alpha” in one response, it stays “Client Alpha” throughout the session. ### What is NOT anonymised Some data is left untouched because it carries no identifying information: * Language codes (en-GB, nl-NL, de-DE, etc.) * Segment counts, word counts, and statistics * Translation status values (draft, translated, approved, etc.) * The actual source and target text you are translating * Technical identifiers (tool names, status labels) ## Enabling Incognito Mode 1. Open **Settings** (gear icon in the Assistant toolbar) 2. Go to the **AI Settings** tab 3. Scroll down to **Context layers (Chat and QuickLauncher)** 4. Tick **Incognito mode** 5. Click **OK** The setting takes effect immediately on your next message. Toggle it off when you no longer need anonymisation. Tip **Tip:** The setting persists across Trados sessions, so remember to toggle it off when you are done sharing. You do not want anonymised data in your regular workflow – it can make the AI’s answers less specific. Note Incognito Mode works with all AI providers, not just Claude. The anonymisation instructions are included in the system prompt that every provider receives. ## Limitations * Incognito Mode instructs the AI to anonymise data in its **responses**. It does not prevent data from being sent to the AI provider – your source text, TM matches, and terminology are still included in the prompt as usual. If you need to prevent data from being sent entirely, disable those context options individually in AI Settings. * The AI does its best to catch all identifying information, but it cannot guarantee 100% coverage. Always review responses before sharing publicly. * [Studio Tools](/trados/ai-assistant/studio-tools/) results (project lists, TM searches, etc.) are also anonymised – the AI receives the real data from the tools but presents it with placeholder names. ## See Also * [Supervertaler](/trados/ai-assistant/) – Overview * [AI Settings](/trados/settings/ai-settings/) – Configure context options and Incognito Mode * [Context layers](/trados/context-layers/) – What context is sent to the AI # Providers The Supervertaler Assistant supports multiple AI providers. You only need one to get started. ## Switching Models The current provider and model are shown in the status area at the bottom of the chat panel. You can switch models in two ways: * **Quick switch** – click the provider/model label directly. A dropdown menu appears with all available models grouped by provider. The current model is marked with a tick. Select a different model to switch instantly. * **Settings** – open the settings dialogue (gear icon) and switch to the **AI Settings** tab for full configuration including API keys, endpoints, and advanced options. ## Supported Providers | Provider | Models | | -------------- | -------------------------------------------------------------------------------- | | **OpenAI** | GPT-5.5, GPT-5.4 Mini | | **Anthropic** | Claude Sonnet 4.6, Claude Haiku 4.5, Claude Opus 4.8 | | **Google** | Gemini 3.1 Flash-Lite, Gemini 2.5 Pro, Gemini 3.1 Pro (Preview), Gemma 4 26B MoE | | **Grok** | Grok 4.3 | | **Mistral** | Mistral Large, Mistral Small | | **DeepSeek** | DeepSeek V4 Pro, DeepSeek V4 Flash | | **OpenRouter** | Claude, GPT, Gemini, DeepSeek, and 200+ others via a single API key | | **Ollama** | TranslateGemma, Qwen 3, Aya Expanse (local, no API key needed) | | **Custom** | Any OpenAI-compatible API endpoint | Note If you want privacy or offline use, try **Ollama** with a local model. No API key or internet connection needed. If you prefer a single account that covers many providers, **OpenRouter** gives you access to 200+ models with one key. ## Choosing a Model For everyday translation questions, a smaller and cheaper model like **GPT-5.4 Mini**, **Claude Haiku 4.5**, or **DeepSeek V4 Flash** works well. For complex tasks like document analysis, prompt generation, or when you need the highest quality suggestions, use a larger model like **Claude Sonnet 4.6**, **GPT-5.5**, or **DeepSeek V4 Pro**. Some features are provider-specific: | Feature | Availability | | ------------------------------------------------------------------------ | --------------------------------- | | [Studio Tools](/trados/ai-assistant/studio-tools/) | All providers except Ollama | | Image attachments | All providers with vision support | | Document attachments | All providers | | [Memory bank](/trados/ai-assistant/super-memory/ai-integration/) context | All providers | ## See Also * [Supervertaler](/trados/ai-assistant/) – Overview * [AI Settings](/trados/settings/ai-settings/) – API keys, endpoints, advanced options * [AI Cost Guide](/trados/ai-cost-guide/) – Token pricing and cost estimates # Studio Tools Studio Tools lets you query your Trados Studio installation using natural language in the Supervertaler Assistant chat. Instead of navigating through menus and dialogs, you can simply ask the assistant about your projects, translation memories, termbases, and project templates – and it will look up the answer for you. Tip Studio Tools works with all major AI providers: **Claude**, **OpenAI**, **Gemini**, **Grok**, and **Mistral**. Only Ollama (local models) does not support tool use and will work as before – plain chat without Trados queries. Caution Studio Tools is under active development and some commands might not yet work as expected. If something doesn’t work, please contact and I’ll get it working as fast as I can. ## How It Works When you send a message in the Supervertaler, the AI automatically decides whether it needs to query Trados Studio to answer your question. If it does, it calls the appropriate tool behind the scenes, reads the result, and presents the information in a clear format. You do not need to use any special syntax or commands. Just ask your question naturally. While a tool is running, the thinking indicator shows what is happening – for example, “Checking Trados projects…” or “Searching translation memory…”. ## Available Tools | Tool | What It Does | | -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ | | **List Projects** | Lists all projects registered in Trados Studio with their name, status, and creation date. Supports filtering by status (in progress, completed, archived) | | **Get Project Details** | Shows detailed information about a specific project, including source and target languages, files, and folder path | | **Project Statistics** | Shows word count and analysis statistics for a project, broken down by match category (perfect, context, exact, fuzzy, new, repetitions) | | **File Status** | Shows the confirmation status of all files in a project – how many segments are not started, draft, translated, approved, or signed off | | **Project Termbases** | Lists termbases attached to a project, with their file paths, enabled/disabled state, and language index mappings | | **TM Info** | Shows detailed information about a specific translation memory, including language pair, segment count, file size, and creation date | | **Search TM** | Searches a translation memory for segments containing specific text. Returns matching source/target pairs so you can see how something was translated before | | **List TMs** | Lists all translation memories found in the Trados Studio TM folders | | **List Project Templates** | Lists all project templates available in Trados Studio | ## Example Questions Here are some things you can try asking in the Assistant chat: ### Projects * “What projects do I have in Trados Studio?” * “Show me all my in-progress projects” * “How many projects do I have?” * “Tell me about the Client Alpha project” * “What languages does the Client Alpha project use?” * “What files are in my latest project?” * “Do I have any completed projects?” * “Which project was created most recently?” ### Project Statistics and Progress * “What are the word counts for the Client Alpha project?” * “Show me the analysis statistics for my project” * “How many new words are in the Client Alpha project?” * “What is the fuzzy match breakdown for this project?” * “What is the translation status of the files in Client Alpha?” * “How many segments are translated vs. not started?” * “Which files still need work?” * “Are any files fully approved?” ### Termbases * “What termbases are attached to the Client Alpha project?” * “Show me the terminology resources for this project” * “Which termbases are enabled?” ### Translation Memories * “What translation memories do I have?” * “List my TMs” * “Tell me about the English-Dutch TM” * “How many segments are in my main TM?” * “How big is my TM?” ### TM Search * “Search my English-Dutch TM for ‘compliance’” * “How was ‘annual report’ translated before?” * “Find segments containing ‘data protection’ in the TM” * “Look up ‘user interface’ in my TM” ### Combined Questions You can combine Studio Tools queries with the assistant’s regular translation capabilities: * “What projects am I working on? And can you also translate this segment?” * “List my projects, then explain the terminology in the current segment” * “Search the TM for ‘privacy policy’ and suggest a translation for the current segment” The assistant handles the tool calls first, then continues with the rest of your question seamlessly. ## Technical Details Studio Tools reads data directly from your local Trados Studio installation. Specifically: * **Projects** are read from the `projects.xml` file in your Documents folder (e.g., `Documents\Studio 2024\Projects\projects.xml`). Project details, statistics, and file status are read from the individual `.sdlproj` files. * **Termbases** are read from the termbase configuration stored in each `.sdlproj` file. * **Translation memories** are found by scanning the `Translation Memories` folder and project folders. TM metadata and search use direct SQLite read-only access to the `.sdltm` files. * **Project templates** are found in the `Project Templates` folder. No data is sent to external services other than the AI provider. The tool results are passed to the AI as part of the conversation so it can format and present them to you. Note Studio Tools currently provides **read-only** access to your Trados data. It cannot create, modify, or delete projects, TMs, or templates. ## See Also * [Supervertaler](/trados/ai-assistant/) – The chat interface where Studio Tools is available * [AI Settings](/trados/settings/ai-settings/) – Configure your AI provider # SuperMemory > >- **SuperMemory** is where you write down the things about a client that you cannot look up: which term they insist on, which wording they rejected last time, how they want dates written, what the previous reviewer changed and why. A translation memory gives the AI your previous wordings and a termbase gives it approved pairs; SuperMemory gives it the *reasoning* – and reasoning is exactly what an AI cannot derive from the source text, so getting it wrong is a real error rather than a stylistic difference. Knowledge lives in **memory banks**: one folder per client, domain, or job, which you switch between from the Supervertaler Assistant toolbar. SuperMemory is one of several [context layers](/trados/context-layers/) the assistant consults, alongside termbases, translation memories, document content and segment metadata. ## A bank is three files Every bank contains the same three Markdown files, plus a folder for source material: | File | What goes in it | | ---------------- | -------------------------------------------------------------------------------------------------------------- | | `brief.md` | Who the client is and anything standing: language pair, register, house preferences, how far to trust the rest | | `terminology.md` | Term decisions, **one table**, one row each | | `style.md` | Prose rules and approved boilerplate – how things are phrased, rather than which term is used | | `reference/` | Source material, unmodified: style guides, PDFs, glossaries. Never sent to the AI | That is the whole structure. You are meant to open these files and edit them by hand – the **📂 Open folder** button in the toolbar is there for exactly that. ### Why terminology is a table A table is the format in which a *wrong* entry is findable. You can scan a hundred rows in half a minute and spot the one that says the wrong thing; you cannot do that with a hundred files. Since the only reason to keep notes is that you can check them, the format that makes checking possible is the one that matters. ```markdown | Source | Target | Scope | Note | |---|---|---|---| | voorkeursvorm | preferred embodiment | domain | Not "preferred form" – term of art | | drager | wearer | project | **Never** "carrier" | ``` **Scope** says how far a row travels – `project`, `client`, or `domain`. It doubles as a promotion queue: a row that turns out to hold for a *second* client belongs in the shared bank. ### `reference/` is the audit trail Everything in the three files is *derived* from something – a style guide, a review round, a conversation. Keeping the original in `reference/` is what lets you check a rule that looks wrong and find out whether it was mis-derived or the source really does say that. Nothing reads this folder automatically, and that is deliberate. ## The `_shared` bank Alongside whichever bank is active, Supervertaler always loads a bank called **`_shared`**. It holds the defaults that are true of *your* work rather than of any one client: house style, domain conventions, jurisdictional rules. **Where they disagree, the active bank wins.** That is what a client bank is for – `_shared` says how you normally translate, and the client says how this one insists on it being done. The AI is told which layer is which, so it can apply the override rather than average the two. A rule earns its place in `_shared` once it has held across more than one client. Until then it stays in the bank where you found it, tagged in the Scope column. Promote by *moving* the row, not copying it, or the two drift apart. **Create `_shared` in the file system, not from the dropdown.** The name is reserved, and the *+ New memory bank…* dialog strips leading underscores from whatever you type – so asking it for `_shared` gets you an ordinary bank called `shared`, which does nothing. Instead use **Open folder**, go up one level to `memory-banks\`, create a folder called `_shared`, and put `brief.md`, `terminology.md` and `style.md` in it. You never have to select `_shared` to use it – it is loaded on top of whichever bank is active. Select it only when you want to edit your defaults. ## Banks ### Where they live ```plaintext D:\Supervertaler\memory-banks\ ├── _shared\ ← always loaded, alongside the active bank │ ├── brief.md │ ├── terminology.md │ └── style.md ├── acme-legal\ │ ├── brief.md │ ├── terminology.md │ ├── style.md │ └── reference\ └── pharma\ └── … ``` ### Switching The **Memory Bank** dropdown lists every bank it finds. Switching is immediate – the next chat turn and the next batch translation both use the new bank, no restart, and your chat history is preserved. The choice persists across Trados sessions. ### Creating Pick **+ New memory bank…** at the bottom of the dropdown, give it a short name (lowercase letters, digits, hyphens, underscores), and it is created with the three files already in place, each carrying its headings and a line explaining what belongs in it. ### Renaming and deleting Not yet available from inside the plugin. Close Trados and rename or delete the folder under `memory-banks\` directly. ## Filling a bank **Write in it.** Open the folder, edit `brief.md`, add rows to the table. This is the normal way, and there is no processing step between what you write and what the AI reads. **[Quick Add](/trados/ai-assistant/super-memory/quick-add/) (Ctrl+Alt+M)** captures a decision without leaving the segment you are on. It appends one row to `terminology.md`. **Drop source material into `reference/`** – a client style guide, a PDF, a glossary. Nothing happens to it automatically. When you want it in the bank, read it and write the parts that matter into the three files, or paste it into the chat and ask the assistant to draft the rows for you to check. That last point is the design in one line: **the AI proposes, you decide what gets written.** ## The Report button **☰ Report** tells you what the bank actually contributes: which files exist and how big, how many rows the terminology table has, **how many tokens get added to a prompt**, whether `_shared` is being applied, and warnings for things that are quietly wrong – a missing brief, a terminology file that is still prose rather than a table, files sitting in the bank root that are never sent to the AI. It reads the files directly, so it is instant and costs nothing. ## Converting a bank from the old layout Banks created before **v18.20.160** used a different structure: seven numbered folders and one file per fact. Those banks are **not read** by the current version – a bank with no `brief.md`, `terminology.md` or `style.md` contributes nothing to a prompt. You will not be left guessing: when the active bank is on the old layout, an amber **⚠ Convert this bank** button appears in the toolbar. Converting folds the old articles into the three files and moves the originals to `reference/_legacy`. **Nothing is deleted.** The conversion copies text across as-is; it does not tidy it up. Expect to read through the result and prune it – particularly the terminology, which arrives as prose and reads far better rewritten as a table. That work is yours on purpose: deciding which of a hundred old decisions still hold is judgement, and a machine that guesses confidently there is how the old system filled up with material nobody could check. ## Why this changed Earlier versions organised a bank as a self-organising wiki: seven folders, one Markdown article per fact, YAML metadata on each, and AI features (Process Inbox, Distill, Health Check) that filed and maintained it for you. It did not survive contact with real use. A single bank reached 136 terminology files – for what is a 136-row table – with a 97-file backlog nobody had processed. About 15% of articles had malformed metadata that silently excluded them from the very filtering the folders existed to enable: they were in the bank, and they were not reaching the AI. Nothing about that was visible from the outside, and by then the bank was far past the point where a human could read it and tell. The lesson was not that automation is bad, but that **knowledge you cannot audit is not knowledge you can rely on**. Three files can be read start to finish in a few minutes. If a rule in there is wrong, you will see it – which is the only mechanism by which it ever gets fixed. Process Inbox, Distill and Health Check are gone. They existed to manage complexity the new structure does not have. ## Working with a bank outside Trados The active bank is exposed over the [Supervertaler MCP server](/trados/mcp-server/), so Claude Desktop, Claude Code or any other MCP client can read it: `get_supermemory_context` for the current picture, `search_supermemory` to look a decision up, `list_supermemory_banks` to see what exists. Access is read-only. Beyond that, a bank is a folder of Markdown files. Edit it in any text editor, search it with ordinary tools, version-control it with Git, keep it in a synced folder so it follows you between machines. [Obsidian](https://obsidian.md/) works nicely if you like it, though with three files per bank you no longer need it to find your way around. Supervertaler Workbench does not use memory banks. That remains a genuine gap rather than a setting you have missed. ## Features | Feature | Description | | ----------------------------------------------------------------------- | ---------------------------------------------------------------- | | [**Quick Add**](/trados/ai-assistant/super-memory/quick-add/) | Capture a term decision while translating (Ctrl+Alt+M) | | [**Active Prompt**](/trados/ai-assistant/super-memory/active-prompt/) | Per-project prompt that Quick Add can also append terminology to | | [**AI Integration**](/trados/ai-assistant/super-memory/ai-integration/) | What gets sent to the AI, and in what order | | [**Obsidian Setup**](/trados/ai-assistant/super-memory/obsidian-setup/) | Optional: installing Obsidian and the Web Clipper | ## Related * [**Context layers**](/trados/context-layers/) – the full menu of context sources, with SuperMemory as one of them * [**AI Settings**](/trados/settings/ai-settings/) – toggles for enabling or disabling SuperMemory context # Active Prompt > Per-project prompt that Quick Add appends terminology to Each Trados project can have an **active prompt** – the prompt that Quick Add appends terminology to. This is also the prompt that is auto-selected in the [Batch Translate](/trados/batch-translate/) dropdown when you open the project. ## Setting the active prompt 1. Open **Settings → Prompts** 2. Right-click any prompt in the tree 3. Choose **Set as active prompt for this project** The active prompt is shown with a pin icon and bold blue text in the Prompt Manager, and a checkmark appears next to its name in the Batch Translate dropdown. The Batch Translate dropdown updates **live** – you do not have to close the Settings dialog for the change to take effect. Cancelling the dialog reverts the change; clicking OK persists it. To clear the active prompt, right-click it again and choose the same menu item (it toggles). Note The active prompt works for any prompt regardless of folder or category. Even if a prompt’s `Category` is not `Translate` (for example, a prompt at the root of the tree with no category set), it will still appear and be pre-selected in the Batch Translate dropdown once you mark it as active. Note The active prompt is saved [per project](/trados/settings/project-settings/). Different Trados projects can have different active prompts. ## See Also * [Quick Add](/trados/ai-assistant/super-memory/quick-add/) * [Batch Translate](/trados/batch-translate/) * [Prompts](/trados/settings/prompts/) * [Per-Project Settings](/trados/settings/project-settings/) # AI Integration > What SuperMemory sends to the AI, and in what order When you translate a segment, run a batch translation, or ask the chat a question, Supervertaler builds the prompt from several context sources. This page covers what SuperMemory contributes. ## What gets sent Two banks, in this order: 1. **`_shared`** – your house defaults, labelled as such. 2. **The active bank** – labelled as overriding the defaults above. Within each, the three files are sent whole: `brief.md`, then `terminology.md`, then `style.md`. There is no selection step and no filtering – whatever is in those files is what the AI sees. `reference/` is never sent. It holds the source material the three files were derived from, and a superseded draft answering as if it were current is precisely the failure it exists to prevent. ## Precedence The prompt states plainly that the client section overrides the house defaults. That instruction only works because the two layers are kept separate rather than merged, so the AI can tell which rule came from where. In practice: `_shared` might say *voorkeursvorm → preferred embodiment*, and a client bank might insist on *preferred form*. The client wins, and the override belongs in that client’s bank – not as an edit to `_shared`, which would change the default for everyone. ## The bank is the selection Earlier versions tried to work out which parts of a bank were relevant: matching your project name against client-profile filenames, detecting the document’s domain, preferring one style guide over another, then loading whichever articles scored highest. None of that happens now. **You pick the bank from the toolbar, and its contents are used** – because you already know which client you are working for, and a detection step could only get that wrong. It also means what reaches the AI is exactly what you would see by opening the folder, with nothing silently excluded. ## Token cost A bank is small enough to send whole. Three files for a single client typically come to a few thousand tokens. The **☰ Report** button tells you the exact figure for the active bank, including what `_shared` adds. Worth checking if you translate in large batches, where the context is re-sent for every call. If a bank does grow past the budget, the shared layer is dropped before the client layer – the client bank was chosen deliberately and overrides the defaults anyway – and terminology is dropped last on each, being the densest content and the hardest for a model to guess. ## How SuperMemory compares with other context sources SuperMemory does not replace your termbases or translation memories – it complements them, adding the reasoning that flat data cannot carry. | Context source | What it provides | What SuperMemory adds | | ----------------------------------------- | ----------------------------------- | ------------------------------------------------------------------------ | | **Termbases** (Supervertaler + MultiTerm) | Term pairs: A = B | The *why*: reasoning, rejected alternatives, client-specific overrides | | **Translation memories** | Previous wordings to anchor style | Rules that hold across segments, and which past work to trust | | **Document content** | What this document is | Conventions and pitfalls the AI cannot read off the page | | **AutoPrompt** | AI-drafted translation instructions | Client and domain context, so the draft starts from what you actually do | For when stacking all of them is or is not optimal, see **Stacking the layers** in [Context layers](/trados/context-layers/#stacking-the-layers). ## Memory-aware chat With SuperMemory context enabled, the chat can answer from your own decisions rather than from general knowledge: * “What register should I use for this client?” * “Does this client prefer *whilst* or *while*?” * “What’s the usual translation for *furtherance* here, and why?” If the bank does not cover it, the assistant falls back to general knowledge and says so. Because the banks are also readable over the [MCP server](/trados/mcp-server/), you can ask the same questions from Claude Desktop or Claude Code while Trados is open. ## Enabling and disabling Toggled in [AI Settings](/trados/settings/ai-settings/): * **Include memory bank in AI context** – for translations and chat. * **Use memory bank when generating prompts (AutoPrompt)** – when AutoPrompt drafts a translation prompt. Both are **off by default**. Turning them off does not delete anything – the files stay on disk. ## See Also * [SuperMemory](/trados/ai-assistant/super-memory/) – how a bank is structured * [Context layers](/trados/context-layers/) – the full menu of context sources * [AI Settings](/trados/settings/ai-settings/) – the toggles * [Batch Translate](/trados/batch-translate/) – batch translation with full context # Obsidian Setup > Setting up Obsidian for your memory banks Memory banks store all knowledge as Markdown files, which you can browse and edit with any text editor. For the best experience, we recommend [Obsidian](https://obsidian.md/) – a free knowledge-base app that visualises the links between your articles as an interactive graph. ## Installing Obsidian 1. Download Obsidian from (available for Windows, Mac, and Linux) 2. Install and open it – choose **Open folder as vault** and select one of your memory bank folders, for example: ```plaintext C:\Users\{you}\Supervertaler\memory-banks\default\ ``` If you keep several banks side by side, you can open each one as its own Obsidian vault and switch between them from Obsidian’s vault switcher. 3. The free version of Obsidian includes everything you need – no subscription required. (The paid Sync and Publish add-ons are not needed.) ## Web Clipper The [Obsidian Web Clipper](https://obsidian.md/clipper) is a free browser extension that lets you clip web pages straight into a memory bank’s `reference/` folder. Install it for Chrome, Firefox, Safari, or Edge. ### Setting up the Web Clipper 1. Install the extension from [obsidian.md/clipper](https://obsidian.md/clipper) 2. Make sure Obsidian is running with the memory bank you want to clip into open as a vault 3. Click the Web Clipper icon in your browser toolbar, then the gear icon (settings) 4. Create a new template (e.g. “memory-bank”) and set: * **Note location:** `reference` * **Vault:** select the memory bank vault you want clippings to land in 5. Optionally add properties: `source_url` = `{{url}}`, `clipped` = `{{date}}` Now when you find a useful reference – a client style guide, a terminology resource, a domain article – click the clipper, hit save, and it lands in that bank’s `reference/` folder. Nothing reads it automatically: `reference/` is source material, so when you want it in the bank, read it and write the parts that matter into `brief.md`, `terminology.md` or `style.md`. Note If you keep several memory banks, create one Web Clipper template per bank so you can choose the destination from the clipper dropdown at clip time. ## Recommended plugins These free Obsidian community plugins enhance the memory bank experience: * **Dataview** – query your vault like a database (e.g. list all terminology articles for a specific client) * **Calendar** – visualise when articles were created or modified * **Graph Analysis** – enhanced graph view with clustering and statistics To install plugins: **Settings → Community plugins → Browse**. ## See Also * [SuperMemory](/trados/ai-assistant/super-memory/) * [User Data Folder](/trados/data-folder/) # Quick Add (Ctrl+Alt+M) > Capture a term decision without leaving the segment You have just worked out how a term should be translated for this client. Quick Add records that decision without breaking your rhythm: it appends one row to the active bank’s `terminology.md` and puts you back in the segment. ## How to use 1. In the Trados editor, select the source text you want to capture. (Optional – the whole source segment is used if you select nothing.) 2. Press **Ctrl+Alt+M**, or right-click and choose **Add to SuperMemory**. 3. Fill in the dialogue: * **Source term** – pre-filled from your selection. The label names your project’s source language. * **Target term** – pre-filled from the target selection if you made one. * **Notes** – why, or when it applies. This is the part that will matter in six months. * **Save as raw note** – puts the text in `reference/` instead of the table, for when the decision is not settled enough to be a row. * **Also append to active translation prompt** – adds the same pair to the TERMINOLOGY table in your [active prompt](/trados/ai-assistant/super-memory/active-prompt/), so it takes effect on the very next Ctrl+T. 4. Click **Add**. The row lands in whichever bank is selected in the toolbar dropdown. To capture into a different bank, switch first. ## What gets written One row, appended to the table in `terminology.md`: ```markdown | fiche | plug | client | Only in the electrical sense; elsewhere "sheet" | ``` Successive additions accumulate in the same table rather than scattering, so the bank stays something you can read start to finish. The **Scope** column is filled in as `client`, which is the safe default – it means “true for this client”. If a decision later turns out to hold for other clients too, move the row to the `_shared` bank so every client gets it. Move rather than copy, or the two drift apart. Note If the bank’s `terminology.md` has no table yet – which is the case for a bank converted from the old layout, where terminology arrives as prose – Quick Add creates one under a **Quick-added terms** heading at the end of the file. Your converted content is left alone. ## Raw notes Tick **Save as raw note** and the text goes to the bank’s `reference/` folder instead, as a plain Markdown note. Use it when the knowledge is not yet a decision: *“fiche can mean either sheet or plug depending on context – check with the client”*. A table row states a rule, and a rule you are not sure about is worse than a note that says you are not sure. Nothing reads `reference/` automatically. When you have settled the question, write the row yourself. ## Review what you capture Quick Add writes what you type, with no AI step in between. That is the point – but it also means a typo in the target term becomes a rule the AI follows. Open `terminology.md` now and again (**📂 Open folder** in the toolbar) and read down the table. That is a two-minute job for a bank of a hundred rows, and it is the whole reason terminology is kept as a table rather than as a folder of files. ## Related * [**SuperMemory**](/trados/ai-assistant/super-memory/) – how a bank is structured * [**Active Prompt**](/trados/ai-assistant/super-memory/active-prompt/) – the prompt Quick Add can also append to * [**Keyboard Shortcuts**](/trados/keyboard-shortcuts/) – the full list # Supervertaler Bridge The **Supervertaler Bridge** is a small localhost-only HTTP service inside the Trados plugin. It is what the [Supervertaler MCP Server](/trados/mcp-server/) talks to: when Claude Desktop, Claude Code or another AI app reads your open project, searches a TM, runs a QA check or drafts translations into the document, every one of those calls goes over this bridge. It runs in the background, exposes its endpoints on `127.0.0.1` only, and is gated behind a per-session bearer token. You never need to configure it. This page exists so you know what is running, why it is safe, and where to look if the MCP server cannot find Trados. ## What it does The bridge exposes the state of the open Trados project to a local client, and accepts a small set of write actions from it: * The active source segment and your current target draft, with a few surrounding segments * TM matches Trados has found for the active segment, and concordance search over the project’s TMs * Termbase hits from your enabled termbases, and term lookups * Project name, file list, languages and statistics * Inserting or updating translations, adding comments, saving the document This is the same context the in-Trados Supervertaler Assistant chat uses for its own answers. The full list of operations is the MCP server’s [tool table](/trados/mcp-server/#what-it-can-do); the bridge is the plumbing under it. ## When it runs The bridge is started automatically when **both** of these are true: 1. You have **Assistant access** – a paid subscription or an active trial. Users without Assistant access never start the bridge. 2. You have **opened the Supervertaler Assistant panel** at least once in this Trados session. The panel is lazy – Trados doesn’t initialise it until you activate it. On start it writes a handshake file, `~/Supervertaler/trados/runtime/bridge.json`, holding the port, the token and the Trados process id; the MCP server reads that file to find the running Studio. The bridge is stopped when Trados Studio exits. If Trados crashes or is force-killed, the next start detects the stale handshake and replaces it with a fresh one. There is no switch to turn the bridge off. Until v18.20.188 a hidden `sidekickBridgeEnabled` setting existed for the retired Supervertaler Workbench integration; it was removed in v18.20.189, since the bridge now exists for the MCP server alone. A user without Assistant access never has a bridge running. ## Privacy and security The bridge is designed to be safe for everyday use: * **Loopback-only.** The HTTP listener binds exclusively to `127.0.0.1`. Other devices on your network – even on the same Wi-Fi – can never reach it. There is a defence-in-depth check that rejects any non-loopback `RemoteEndPoint` even if the binding ever drifts. * **Per-session authentication token.** A fresh GUID is generated every time the bridge starts. Clients must present it as a `Bearer` token. Stale tokens from previous sessions are useless. * **Random high port.** The bridge picks a random port in the 49152–65535 range to avoid collisions with other local services. * **No external network access.** The bridge only listens; it never reaches out to any external service. ## Troubleshooting The bridge writes a diagnostic log to two locations on every start: * `~/Supervertaler/trados/runtime/bridge.log` – under your Supervertaler user-data folder * `%TEMP%\Supervertaler-bridge.log` – guaranteed-writable fallback The log is truncated on every plugin start, so it always reflects the current Trados session. The first lines record the resolved data-folder path so you can see exactly where the plugin is looking. Useful entries to look for: | Log line | Meaning | | ---------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- | | `Initialize: HasAssistantAccess=false` | Your licence isn’t picked up as paid or trial. The bridge is correctly skipped in this case. | | `port NNNNN bind failed: HttpListenerException code=5` | Windows is refusing to let the plugin bind to a localhost port. Rare; usually means a strict group-policy environment. | | `Start() complete. Bridge live on http://127.0.0.1:NNNNN/` | All good – the bridge is running and the handshake file should be at `~/Supervertaler/trados/runtime/bridge.json`. | If `bridge.json` exists and contains `port`, `token`, `pid`, and `startedAt`, the bridge is healthy. The MCP server’s `list_trados_instances` tool shows every running Studio it can see. ## Endpoint reference (advanced) For developers who want to integrate other tools with the bridge, here are the two oldest endpoints; the MCP server’s tool definitions (`mcp-tools.json` in the plugin) describe the rest, each with its method, path and parameters. **The URL prefix is versioned** so future schema changes can ship without breaking older clients. ### `GET /v1/active-context` Returns a JSON snapshot of the current Trados project state. Authentication via `Authorization: Bearer ` from the handshake file. ```json { "available": true, "project": { "name": "ACME-PROJ-001", "fileName": "20260101 PROJ-001 Application as filed.docx", "sourceLang": "nl-BE", "targetLang": "en-US" }, "activeSegment": { "source": "...", "target": "..." }, "surroundingSegments": [ { "source": "...", "target": "..." } ], "tmMatches": [ { "score": 95, "source": "...", "target": "...", "tmName": "..." } ], "termbaseHits": [ { "source": "...", "target": "...", "termbaseName": "...", "definition": "...", "domain": "...", "notes": "..." } ] } ``` When no document is active, returns `{"available": false}` with HTTP 200. ### `POST /v1/insert-translation` Inserts text into the active Trados target segment via the same code path as the in-Chat Apply-To-Target button. Request body: ```json { "text": "The translation to insert" } ``` Response on success: ```json { "ok": true } ``` Response on failure (e.g. no active segment): ```json { "ok": false, "error": "no active document" } ``` ## Related pages * [Supervertaler MCP Server](/trados/mcp-server/) – the client this bridge exists for * [Supervertaler](/trados/ai-assistant/) – the in-Trados chat that uses the same context fields the bridge exposes * [User Data Folder](/trados/data-folder/) – where the handshake file lives # AI Cost Guide This page explains how AI costs work in Supervertaler for Trados and how to keep them under control. It deliberately avoids quoting exact per-model prices – those change often, and Supervertaler already shows you the **real, current cost** of every operation in the **Reports** tab. For exact figures, see [Estimates vs actual cost](#estimates-vs-actual-cost) below. Note AI provider costs are **separate** from your Supervertaler licence. You pay the AI provider directly for the tokens your requests consume. Supervertaler does not add any markup. ### Estimates vs actual cost The **Reports** tab shows **real billed token counts and cost** as reported by your provider’s API – with cache-hit tokens broken out (e.g. `830,000 in (720,000 cached) / 32,000 out · $1.36`). When the cost has no `~` prefix, that’s the actual amount your provider will charge. **This is the single best place to see what you’re actually spending** – it’s live, per-operation, and provider-reported. The number is cache-aware: * **Anthropic (Claude) native + OpenRouter → Claude**: real usage from `usage.input_tokens` + `cache_creation_input_tokens` + `cache_read_input_tokens`. Cache reads are billed at 0.1× the input rate, cache writes at 1.25×. * **OpenAI**: real `prompt_tokens` and `completion_tokens` plus `prompt_tokens_details.cached_tokens` for the auto-cache discount (50% off cached input). * **DeepSeek**: real `prompt_tokens` / `completion_tokens` with auto-cache (90% off cached input). * **Gemini 2.5+**: real `usageMetadata` with implicit cache (75% off cached input). For these providers the in-app number is the authoritative billable figure (modulo any account-level credits or monthly minimums you may have). The chars/4 estimate is still used as a fallback when the provider didn’t return usage info – this affects Ollama (local, free anyway), some provider edge cases, and any response shape we couldn’t parse. In those cases the cost still appears with a `~` prefix to flag it as an estimate. If you want to cross-check against your provider’s own dashboard: | Provider | Where to look | | ---------------------- | --------------------------------------------------------------------------------------- | | **Anthropic (Claude)** | [platform.claude.com – Cost](https://platform.claude.com/workspaces/default/cost) | | **OpenAI (GPT)** | [platform.openai.com – Usage](https://platform.openai.com/settings/organization/usage) | | **Google (Gemini)** | [console.cloud.google.com – Billing reports](https://console.cloud.google.com/billing/) | | **xAI (Grok)** | [console.x.ai – Usage](https://console.x.ai/team/default/usage) | | **Mistral AI** | [console.mistral.ai – Usage](https://console.mistral.ai/usage) | | **OpenRouter** | [openrouter.ai – Activity](https://openrouter.ai/activity) | | **DeepSeek** | [platform.deepseek.com – Usage](https://platform.deepseek.com/usage) | | **Ollama** | Free – local execution, no provider console. | The in-app number and the provider dashboard should agree to within rounding for any given run. If you see a meaningful gap, the most likely causes (in order) are: a provider-side credit or volume discount the in-app calculator can’t see; the in-app pricing table being a little behind a recent rate change; or, for fallback (estimate) cases, the chars/4 heuristic over- or under-counting tokens for that particular language and content type. ### How costs are calculated AI providers charge per **token** – a unit of text roughly equal to ¾ of a word. Costs depend on: * **Input tokens** – the text you send (source segment, system prompt, terminology context) * **Output tokens** – the text the model returns (translated segment, proofread text, generated prompt) Because Supervertaler translates **segment by segment**, the system prompt and terminology context are included with every segment. For a typical 5,000-word document (\~250 segments), the token usage works out roughly like this: | Task | Input tokens | Output tokens | | ------------------- | ------------ | ------------- | | **Batch Translate** | \~125,000 | \~8,000 | | **AI Proofreader** | \~140,000 | \~8,000 | | **AutoPrompt** | \~10,000 | \~2,000 | These are estimates for a representative document – actual usage varies with segment length, terminology context size, and prompt complexity. Token counts like these are fairly stable; what changes is the **price per token**, which is why this guide points you to live figures rather than quoting them. ### How much will it cost? There’s a wide spread between models. As a rough mental model: * **Local models (Ollama)** are **free** – they run on your own computer, with no API charges at all. The trade-off is that quality depends on your hardware, and they’re generally less capable than cloud-hosted models. If you have a computer with 8+ GB of RAM, TranslateGemma 12B delivers surprisingly good results for free. * **Budget cloud models** – the “Mini”, “Flash-Lite” and “Small” tier from each provider (e.g. GPT-5.4 Mini, Gemini 3.1 Flash-Lite, Mistral Small, Claude Haiku 4.5) – typically cost a small fraction of a cent per segment. They’re excellent for routine, high-volume translation. * **Flagship models** – Claude Opus 4.8, GPT-5.5, Gemini 3.1 Pro and the like – can run roughly 10–50× the price of the budget tier. Reserve them for specialised content where the quality difference earns its keep. To see what a model **actually** costs for your work, run one operation and check the **Reports** tab – it shows the real billed cost. For OpenRouter, expect the underlying provider’s rate plus a small platform fee. ### Our recommendation Tip **If you could only pick one model for everything – translation, proofreading, and chat – we would recommend Claude Sonnet 4.6.** It follows translation instructions precisely, handles terminology constraints well, is fast enough for batch operations, and delivers consistently high quality across legal, technical, and general content – at a cost that works out to a small fraction of a cent per segment. For budget-conscious batch work, **GPT-5.4 Mini** or **Gemini 3.1 Flash-Lite** offer excellent quality at a fraction of the price. For the absolute highest quality on specialised content, **Claude Opus 4.8** or **GPT-5.5** are worth the premium. ### Token pricing Supervertaler’s in-app cost figures come from a built-in per-token pricing table. Because provider prices change regularly, that table is occasionally a little behind a recent rate change – the **Reports** tab’s provider-reported figures are always the authoritative ones. For the definitive current rates, check the provider’s own pricing page: [OpenAI](https://openai.com/api/pricing/) · [Anthropic](https://www.anthropic.com/pricing#anthropic-api) · [Google Gemini](https://ai.google.dev/gemini-api/docs/pricing) · [xAI](https://docs.x.ai/developers/models) · [Mistral](https://mistral.ai/technology/) · [DeepSeek](https://api-docs.deepseek.com/quick_start/pricing/) · [OpenRouter](https://openrouter.ai/models) ### Tips for managing costs * **Start with a budget model** – GPT-5.4 Mini, Gemini 3.1 Flash-Lite, or Mistral Small are excellent for routine translation at a fraction of the cost of a flagship. * **Use premium models selectively** – reserve GPT-5.5, Claude Opus 4.8, or Gemini 2.5 Pro for specialised content (legal, medical, patents) where the quality difference justifies the cost. * **Try Ollama for zero cost** – if you have a computer with 8+ GB of RAM, TranslateGemma 12B delivers surprisingly good results for free. * **Check your usage** – the **Reports** tab lists every AI call live with its token count and cost; the **[Token Usage & Costs](/trados/usage-costs/)** report totals your spend over time (by project, client, model or month) and exports it to CSV/Excel; and your provider’s own console (see the [Estimates vs actual cost](#estimates-vs-actual-cost) table above) shows the authoritative billable figure. * **Set a monthly budget** – give Supervertaler a soft monthly limit (Settings → AI Settings) and it will warn you before a batch once you’ve reached it. See [Token Usage & Costs](/trados/usage-costs/#monthly-budget). ### Built-in cost protection Supervertaler includes several safeguards to help you avoid unexpected costs: #### QuickLauncher prompts are standalone When you run a prompt from the QuickLauncher menu (Alt+Q), only the prompt itself is sent to the AI – **not the chat history**. This means a simple terminology query costs only what it needs to, even if you have a long conversation in the chat window. #### Chat token budget Regular chat messages include recent conversation history so the AI can follow your discussion. However, Supervertaler automatically trims older messages when the history grows too large (\~50,000 tokens). This prevents costs from spiralling when previous messages contained large context blocks (e.g. full document content). #### Cost warning If a request is estimated to cost more than $0.50 in input tokens, a confirmation dialogue appears showing the estimated token count and cost. You can cancel before the expensive request is sent. ![]() Note **Keep an eye on the cost indicators.** Every AI response in the chat shows the estimated token count and cost. You can also review all prompts and their costs in the **Reports** tab. #### Choosing the right model For everyday work – chat queries, terminology questions, QuickLauncher prompts – use **GPT-5.4 Mini** or another budget model. Reserve premium models like **GPT-5.5** or **Claude Opus 4.8** for AutoPrompt and complex tasks where the quality difference justifies the cost. ### See also * [Token Usage & Costs](/trados/usage-costs/) – the persistent usage log, the Usage & Costs report, CSV/Excel export, and the monthly budget * [AI Settings](/trados/settings/ai-settings/) – configure your API keys and choose a model * [Batch Translate](/trados/batch-translate/) – translate segments in bulk * [AI Proofreader](/trados/ai-proofreader/) – proofread translated segments * [AutoPrompt](/trados/generate-prompt/) – generate translation prompts * [Licensing & Pricing](/trados/licensing/) – Supervertaler subscription plans # Ai Proofreader The AI Proofreader checks your translated segments for errors using AI. It identifies issues such as mistranslations, omissions, grammar problems, and inconsistencies, and presents the results as clickable issue cards in the **Reports** tab. ## Starting a Proofreading Run 1. Open the **Supervertaler Assistant** panel (View > Supervertaler Assistant) 2. Switch to the **Batch Operations** tab 3. Select **Proofread** (instead of Translate) 4. Choose a **scope** from the dropdown 5. Optionally select a proofreading **prompt** from the prompt selector 6. Click **Proofread** ## Scope The scope dropdown controls which segments are checked: | Scope | Description | | ------------------------------------ | ------------------------------------------------------------------- | | **Translated only** | Checks only segments with Translated status | | **Translated + approved/signed-off** | Checks segments with Translated, Approved, or Signed-off status | | **All segments** | Checks every segment that has target text | | **Filtered segments** | Checks only segments visible after applying a Trados display filter | | **Filtered (translated only)** | Checks translated segments within the current filter | ## Prompt Selection When in Proofread mode, the prompt dropdown shows only prompts with the **Proofread** category. This keeps the list focused – translation prompts are hidden. If no prompt is selected, the AI uses a default proofreading instruction that checks for accuracy, completeness, grammar, and consistency. Note You can create custom proofreading prompts in the [Prompt Manager](/trados/settings/prompts/). Set the category to **Proofread** so they appear in the dropdown when proofreading. ### Default Proofreading Prompt The shipped **Default Proofreading Prompt** is deliberately slim. The hardcoded base of every Batch Proofread already includes the persona, the five quality categories (accuracy / completeness / terminology / grammar / number formatting), the output format, the “no full corrected translations” rule, and language-specific checks for Dutch, German, and French – so the prompt itself only needs to add what’s missing from the base. What the default prompt *does* add: * **Default to OK.** Raise an ISSUE only when a specific, demonstrable problem can be pointed to in the translation – never speculative or hypothetical concerns. * **Citation discipline.** When flagging a terminology consistency issue, the AI must cite specific source segment numbers in the **Evidence:** field – e.g. *“`'trekriem'` rendered as `'pull belt'` in `[SEGMENT 0031]`, `'draw strap'` in `[SEGMENT 0084]`”*. Inconsistency claims without concrete citations are not allowed. * **Source query distinction.** If the source itself contains an error (typo, duplication, missing word, internal inconsistency), the AI prefixes the Issue line with **“Source query:”** and notes whether the translation handled it correctly. A faithful rendering of a flawed source is not a translation error. * **Explicit boundaries.** The AI does not re-engineer the source, propose alternative terminology without a citation, flag stylistic preferences as errors, or flag empty target lines (those mean the segment hasn’t been translated yet). Note **If you want to customise:** clone the default in the [Prompt Manager](/trados/settings/prompts/) and edit your copy. The default itself is read-only and gets refreshed when the plugin updates. Your clone keeps all your changes. ## Reports Tab Proofreading results appear in the **Reports** tab of the Supervertaler Assistant panel. Each issue is shown as a clickable card containing: * **Segment number** – the actual per-file segment number as shown in the Trados editor grid * **Issue description** – what the AI found wrong * **Evidence** *(when applicable)* – specific source segment numbers the AI cites to back up the claim, shown in italic grey between the issue and the suggestion. Required for terminology consistency claims (see *Default Proofreading Prompt* below) so you can verify the inconsistency yourself by jumping to the cited segments. * **Suggestion** – the AI’s recommended fix (if available) Right-click any card to copy the issue, the evidence, the suggestion, or the whole card to the clipboard. ### Navigating to Issues Click any issue card to navigate directly to that segment in the Trados editor. This works correctly in multi-file projects – the plugin uses the segment’s internal identifiers to find the exact segment. ### Dismissing Issues Each issue card has a checkbox. Tick it to dismiss the issue and remove it from the list. This lets you work through the results one by one, keeping track of which issues you have already addressed. When all issues have been dismissed, the Reports tab shows “All issues addressed – well done!” ### Saving the report (from v18.20.187) Every completed proofreading run is written to disk automatically, as a Markdown file in `trados\ eports` inside your Supervertaler data folder – one section per segment with the issue, suggestion, evidence, source and target. The footer of the Reports tab names the file (hover it for the full path), and the Batch Operations log shows the path too. So a report that has been cleared, or lost when Studio refreshes the panel, is never gone. To keep a copy somewhere else – a client folder, an email attachment – click **Save report…** next to **Clear**. It writes the same Markdown to a location of your choice, and it is only enabled while a proofreading report is showing. ### The report stays until you are finished (from v18.20.187) The report also stays in the Reports tab until you clear it. Close Studio and reopen the document, or switch to another document and back, and the report is put back as you left it: the issues you had ticked off stay ticked off, and clicking a card still takes you to the segment. A report is only ever restored for the document it was made on. Press **Clear** when you are done with it. ### Clearing Results Click the **Clear** button at the top of the Reports tab to remove all results and start fresh. The automatic copy on disk is unaffected. ### Run Summary After a proofreading run, the Reports tab shows: * Total number of issues found and segments checked * Run timestamp and duration in the footer ## Adding Issues as Trados Comments Check the **“Also add issues as Trados comments”** checkbox in the Batch Operations tab (visible only in Proofread mode) before starting the run. When enabled, each issue found by the proofreader is also inserted as a Trados segment comment, so you can see the issues directly in the editor without switching to the Reports tab. ## AI Context in Proofreading Batch Proofread builds a richer context than Batch Translate – it has to, because verifying whether a term is rendered consistently across the document is exactly the kind of question the AI needs the whole document to answer. * **Full bilingual document context** – when **Include document context** is enabled in [AI Settings](/trados/settings/ai-settings/), every segment in the document is included with both source AND target text, with no truncation. This is what makes target-side consistency verifiable: the AI can check “this term is rendered as X in \[SEGMENT 0031] and Y in \[SEGMENT 0084]” against the actual document, not against a guess. Segment numbers in the document context match the `[SEGMENT XXXX]` numbers the AI sees in the batch it’s reviewing, so citations cross-reference both ways. * **Termbase terms** – terminology from enabled termbases is checked against the translations, including term definitions and domains when that option is enabled. Forbidden terms are flagged with a `⚠️ DO NOT USE` marker so the AI knows to flag them as issues if it sees them in the translation. * **Language-specific checks** – Dutch, German, and French targets get auto-included quality checks (compound spelling, dt-errors, de/het articles for Dutch; capitalisation and case system for German; accents and punctuation spacing for French). These come from the hardcoded base and don’t need to be in your custom prompt. * **Custom prompts** – the selected proofreading prompt provides domain-specific quality checks on top of all of the above. TM matches and surrounding segments are **not** included in proofreading – these are Chat & QuickLauncher features only. See the [AI Settings](/trados/settings/ai-settings/) page for a full comparison table. Caution **Token cost:** Sending the full bilingual document roughly doubles the context size compared to source-only. For typical patent / legal / technical jobs (under \~500 segments) this is a minor cost increase – usually a few extra cents per batch on Sonnet-class models. For very long documents the cost scales linearly; if you proofread a 5,000-segment book you may want to disable Include document context and rely on per-batch context only. ## Clipboard Mode If you prefer to use a web-based AI (ChatGPT, Claude, Gemini, etc.) instead of an API, tick the **Clipboard Mode** checkbox. Supervertaler builds a complete proofreading prompt with both source and target text for each segment and copies it to your clipboard. See [Clipboard Mode](/trados/clipboard-mode/) for full details. ## Tips ### Start with Confirmed Segments Use the **Confirmed Only** scope to check segments you consider finished. This avoids noise from segments that are still being worked on. ### Use Domain-Specific Proofreading Prompts Create custom proofreading prompts tailored to your domain. For example, a medical proofreading prompt can check for correct use of clinical terminology, while a legal proofreading prompt can verify that defined terms are used consistently. ### Review After AI Translation The AI Proofreader pairs well with [Batch Translate](/trados/batch-translate/). After translating a batch of segments with AI, run the proofreader to catch any issues before final review. ### Combine with Display Filters Use Trados display filters to isolate specific segments (e.g., segments containing a certain term), then proofread only those filtered segments for targeted quality checks. *** ## See Also * [Clipboard Mode](/trados/clipboard-mode/) * [Batch Translate](/trados/batch-translate/) * [Prompts](/trados/settings/prompts/) * [Supervertaler](/trados/ai-assistant/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) # AutoTagger AutoTagger looks at where the inline tags sit in the **source** segment and inserts that same set of tags into your **existing translation** at the right places – without changing any of the translated words. It is for the common case where a target has the correct translation but is **missing its tags or has them in the wrong spots** – typically after machine translation, pasting from another tool, or typing the target by hand. In Trados Studio those segments otherwise trip the tag QA checks even though the wording is fine. ## How to use it With an active segment that has a correct translation but wrong/missing tags: * Right-click in the editor → **Auto-tag active segment**, or * Press **Ctrl+Alt+G**.  Trados Undo (`Ctrl+Z`) reverts it. AutoTagger works **without opening the Supervertaler Assistant pane**, and it never opens that pane – it reads the active segment from the editor and its settings from disk, so it just works without disturbing your layout, even straight after a Trados restart. ## How it works 1. AutoTagger reads the source segment’s inline tags and your current target text. 2. It asks the AI to place that exact set of tags into your translation at the correct positions. 3. It **validates** the result before writing: the tag set must match the source, the words must be unchanged, and the tags must be well-formed. 4. It re-inserts the tags into your **exact** target, so punctuation such as curly quotes is preserved verbatim. 5. If the AI’s output doesn’t validate it **retries once**, and otherwise leaves the segment **untouched** – so it never writes broken tags. AutoTagger reuses the same tag engine as Batch Translate, so the tags it writes are real Trados inline tags, not placeholders. ## Configuring it Settings → **Prompts** → **AutoTagger Instruction**. This editable field tells the AI how to place the tags. It supports these placeholders: | Placeholder | Meaning | | ----------------- | ---------------------------------------- | | `{{SOURCE_TEXT}}` | The source segment, with its inline tags | | `{{TARGET_TEXT}}` | Your current translation (tags stripped) | | `{{TAG_LIST}}` | The list of tags that must be placed | ## Tracking cost AutoTagger’s AI calls are logged under their own **“AutoTagger”** task in [Token Usage & Costs](/trados/usage-costs/), with token counts and cost like every other AI call. ## Notes * **v1 is single-segment.** A batch mode may follow. * **Shortcut:** Ctrl+Alt+G triggers AutoTagger. The floating TermLens popup opens with **Alt+L**; you can reassign it in Trados’ keyboard settings if you like. * AutoTagger mirrors the [AutoTagger feature in Supervertaler Workbench](/workbench/ai-translation/autotagger/). *** ## See Also * [Batch Translate](/trados/batch-translate/) * [Import/Export](/trados/import-export/) * [Token Usage & Costs](/trados/usage-costs/) * [Keyboard Shortcuts (Trados)](/trados/keyboard-shortcuts/) # Batch Operations The **Batch Operations** tab in the Supervertaler Assistant panel provides two AI-powered modes for processing multiple segments at once: | Mode | Description | | ----------------------------------------------- | ---------------------------------------------------------------- | | **[Batch Translate](/trados/batch-translate/)** | Translate segments using AI with customisable prompts | | **[AI Proofreader](/trados/ai-proofreader/)** | Check translations for errors, inconsistencies, and style issues | Switch between modes using the **Mode** dropdown at the top of the Batch Operations tab. Both modes share the same prompt selector, provider/model configuration, and scope options. Prompts are filtered by mode – Translate prompts appear in Translate mode, Proofread prompts appear in Proofread mode. You can click the **provider/model label** to quickly switch AI models via a flyout menu – the same menu available in the Chat tab. ### Clipboard Mode Both Translate and Proofread modes support **[Clipboard Mode](/trados/clipboard-mode/)** – an alternative workflow that lets you use any web-based AI (ChatGPT, Claude, Gemini, etc.) without an API key. Tick the **Clipboard Mode** checkbox to switch from API-based processing to a manual copy/paste workflow. See [Clipboard Mode](/trados/clipboard-mode/) for full details. ### Preview prompt Next to the action button, the **👁 Preview prompt** link opens a read-only dialog showing **exactly what would be sent to the AI** for the current configuration: the assembled system prompt (including the active custom prompt, termbase entries, language-specific checks, and the full bilingual document context for proofread), followed by the numbered segment list. No LLM call is made. This is useful for: * **Sanity-checking before an expensive call** – see what the model will actually receive (including how many tokens of context, whether your termbase is being included, whether the right segments are in scope) before clicking Translate / Proofread. * **Debugging unexpected output** – if the AI produces an odd suggestion, the preview shows you the exact prompt the model was answering, so you can see whether the issue is in your custom prompt, the termbase, the document context, or the segment list. * **Manually pasting into a web LLM** – the dialog has its own *Copy to clipboard* button, so you can use it as a one-shot “send this to ChatGPT/Claude/Gemini” path without toggling Clipboard Mode. The preview works in both **API mode** and **Clipboard Mode** without switching, and is available for both Translate and Proofread. ### SuperBench Next to Preview prompt, the **⚖ SuperBench…** link (from v18.20.187) translates the first segments of the document with three models under exactly these batch settings and has a judge compare them blind, with a recommendation for this project. Nothing is written to the document. See [SuperBench](/trados/superbench/). ### List numbering as structure context Word’s claim numbers, lettered steps and bullets are not segment text, so the AI used to never see them. Since v18.20.188 (always on from v18.20.189; see [AI Settings](/trados/settings/ai-settings/#list-numbering-as-structure-context-from-v1820188-always-on-from-v1820189)), Batch Translate prefixes the first segment of each numbered paragraph with its marker inside a sentinel – `[#e)]`, `[#9.]` – and tells the model it is structure, never to be reproduced. The log reports how many markers were found, and any marker the model echoes back is removed before the target is written. Preview prompt shows the markers as they will be sent. ### Reference numbers Technical documents – patents, manuals, specifications – point at parts of a drawing with numbers in brackets: “the valve (12) is connected to the pipe (3a)”. The **Reference numbers - like “(12)”** link lists every such number in the open document, how often it is used, and the sentence that first mentions it. It also names any number the sequence skips – a document citing (1) to (4), (6) and (7) is told that (5) is cited nowhere, which usually means a part was renumbered or dropped while the document was drafted. It reads the whole document regardless of the Scope setting, makes no AI call, and opens the list in the Chat tab. A document without such numbers simply says so. Useful for spotting a number that is cited once and never explained, or a part that changes number halfway through. ### FigureLens (from v18.20.189) The AI sees the text of your documents, not the pictures in them. **FigureLens** fixes that in two steps, and it tells you at each step what will happen and what it costs. The **FigureLens…** link opens it.  Click 1: the **FigureLens…** link, under Batch Operations.  Clicks 2 and 3, in the panel that opens: **Extract images to a folder** (free, no AI), then **Describe images with AI** – one AI request per image. **Describe from the text only** is the free alternative to that second button, not a third step.  What is in the document: a picture the AI cannot see.  What Step 2 writes: the description the AI reads with every request. **Your documents** – a summary line and a two-column table of every Word document in the project, name in one column and what was found in it in the other (the source-language files in Studio’s Files view; reference files are skipped): how many images each holds and how many carry a figure label, the documents with images first, then those with none in grey, then any that could not be read in red. Nothing outside the project is scanned; if the pictures are in a separate document, add it to the project. **Step 1 – Extract images to a folder…** The first time, it asks where to put the images; a new, empty folder next to the job is fine, and the choice is remembered for this project. Then it copies the images out of the documents into that folder, named after their figure numbers (`Figure 01.png`, `Figure 02.png`…). If more than one document has images, each gets its own sub-folder inside it, so two documents’ `Figure 01` cannot overwrite each other. Drawings held as vector images (EMF or WMF, which is what a drawing placed from CAD usually is) are rendered to PNG on the way out, because no AI can read a metafile. Free, no AI. If you already keep the images in a folder of your own, use **Already have the images in a folder? Choose it…** instead, and later **Change…** to switch folder or **Open folder** to look at them. **Step 2 – Describe images with AI.** Shows each image to the AI together with what the document says about it, and saves a description of each – what it shows, which figure it is, which parts and numbers appear in it – as `figures.md` in the active memory bank. That file is read by the AI with every request from then on, so it knows what your images mean in this document. Each document gets its own table in the file. One AI request per image; the button says how many and to which provider before you click. It asks before replacing descriptions that already exist. **Describe from the text only** is the free alternative: only what the document itself says about each figure, without looking at the images. Use one or the other. **Result** – whether the descriptions exist yet, when they were saved, how many figures, and whether they came from the AI or from the text alone. The descriptions are saved in the active memory bank, and that must be a bank of this project’s own: the shared bank is read by every project, so the panel refuses to write there and instead offers **Create memory bank “” for this project and switch to it**, one click. If you already keep a bank for the project, pick it on the Chat tab first. Buttons that cannot run yet say why: no images in the documents, step 1 not done, or no active memory bank. The **Document images report** link at the bottom opens the full per-image listing in the Chat tab. **Reference numbers** stays a separate link above, because it is a check on the text, not part of this pipeline. ### AutoPrompt The Batch Operations tab also includes an **[AutoPrompt](/trados/generate-prompt/)** link that uses AI to create a comprehensive, domain-specific translation prompt based on your project’s content, terminology, and TM data. ## See Also * [Clipboard Mode](/trados/clipboard-mode/) * [SuperBench](/trados/superbench/) * [AutoPrompt](/trados/generate-prompt/) * [Prompts](/trados/settings/prompts/) * [AI Settings](/trados/settings/ai-settings/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) # Batch Translate Batch Translate lets you translate multiple segments at once using AI. It is located in the **Supervertaler Assistant** panel, on the **Batch Operations** tab. Note The Batch Operations tab also supports **Proofread** mode for AI-powered quality checking. See [AI Proofreader](/trados/ai-proofreader/) for details. ![]() ### Starting a Batch Translation 1. Open the **Supervertaler Assistant** panel (View > Supervertaler Assistant) 2. Switch to the **Batch Translate** tab 3. Choose a **scope** from the dropdown 4. Choose a **prompt** from the prompt selector 5. Click **Translate** ### Scope The scope dropdown controls which segments are translated: | Scope | Description | | --------------------------- | -------------------------------------------------------------------------------------------------------------- | | **Empty Segments Only** | Translates segments that have no target text | | **All Segments** | Translates every segment in the file | | **All unfinished segments** | Translates every segment that isn’t Translated, Approved or Signed off – drafts and rejected segments included | | **Filtered Segments** | Translates only the segments currently visible after applying a filter | | **Filtered Empty Only** | Translates empty segments within the current filter | Note **Locked segments are never sent to the AI**, in any scope – batch translation and batch proofreading both skip them, and the segment counters above the Translate button exclude them. *(From v18.20.154; earlier versions incorrectly included locked segments in batch runs.)* ### Prompt Selection Choose a prompt to guide the AI translation style and domain. The prompt selector shows: * **Default Translation Prompt** – a general-purpose prompt that works well for most content types. Use it as-is or duplicate it in the Prompt Manager and customise it for your domain. * **Custom prompts** – your own prompts created in the Prompt Manager The **active prompt** for the current project is marked with a checkmark in the dropdown. When you open a project that has an active prompt set, it is automatically selected. See [Memory banks – Active Prompt](/trados/ai-assistant/super-memory/active-prompt/) for how to set the active prompt. Tip **Tip:** If you save a prompt with the same name as your Trados project, the dropdown will auto-select it whenever you open that project. For example, a prompt called “HAYNESPRO” will be auto-selected when working in a project called HAYNESPRO. Note For specialised fields (medical, legal, patent, etc.), create a custom prompt with domain-specific terminology rules and instructions. A tailored prompt is the single most effective way to improve translation quality. ### Provider and Model The current AI provider and model are displayed below the prompt selector. Click the provider/model label to open a flyout menu where you can switch models instantly – the same menu available in the Chat tab. Alternatively, open the settings dialogue (gear icon in the TermLens header) and go to the **AI Settings** tab. Your **custom OpenAI-compatible profiles** (institutional gateways, self-hosted endpoints) appear at the bottom of the same menu under *Custom (OpenAI-compatible)*, each showing its configured model, with the active profile ticked – so switching between custom endpoints is one click, no trip to Settings. *(From v18.20.154.)* ### Progress and Logging During translation: * A **progress bar** shows overall completion * A **real-time log** displays the status of each segment as it is translated * The **Stop** button aborts the batch at any time – segments already translated are kept ### Translate Active Segment (Alt+T) Press **Alt+T** to translate the active segment instantly. This uses the same provider, model, and prompt as Batch Translate, so you can switch prompts or providers and immediately use them for single segments with Alt+T. Alt+T is also available via right-click in the editor (“Translate active segment”). Note Older installs may still have this on `Ctrl+T` – the default moved to `Alt+T` in v18/19.20.119 because `Ctrl+T` collides with a Trados factory shortcut. See [Keyboard shortcuts](/trados/keyboard-shortcuts/) for how to reassign it. #### How it works 1. The active segment’s source text is sent to the AI provider configured in AI Settings 2. The selected prompt (from the Batch Translate tab) is applied, along with termbase terms 3. The same document context and SuperMemory context as a batch run is included *(from v18/19.20.149 – earlier versions sent the segment on its own, which is why single segments used to translate noticeably worse than a batch)* 4. The translation is written directly into the target cell 5. Inline tags (bold, italic, field codes, etc.) are preserved in the translation In short: a single segment now gets everything a batch run gets, minus the other segments – same prompt, same terminology, same document context. On versions before 18/19.20.149, batch translating even small ranges gave better results than going segment by segment; from 18/19.20.149 the two are equivalent. ### AI Context in Batch Translate Batch Translate uses several context sources from your [AI Settings](/trados/settings/ai-settings/) to improve translation quality: * **Document content** – when enabled, all source segments are included in the system prompt so the AI can determine the document type (legal, medical, technical, etc.) and adapt its style accordingly. This is shared across all batches. * **Termbase terms** – terminology from enabled termbases is injected into the prompt, including term definitions and domains when that option is enabled. * **Custom prompts** – the selected prompt provides domain-specific translation instructions. TM matches and surrounding segments are **not** included in Batch Translate – these are Chat & QuickLauncher features only. See the [AI Settings](/trados/settings/ai-settings/) page for a full comparison table. ### Backup TMX The **Auto-backup translations to TMX** checkbox is ticked by default. When enabled, Supervertaler writes every translated segment to a TMX file as it arrives from the AI. If Trados crashes mid-run, you can recover the completed translations without re-running the batch. The TMX files are also useful outside of crash recovery – you can import them into any TM in Trados, memoQ, Wordfast, or any other CAT tool that accepts standard TMX. Click **Open folder…** next to the checkbox to open the backup folder directly in Windows Explorer. To disable backups for a particular run, simply untick the checkbox before clicking Translate. #### How it works * A new `.tmx` file is created at the start of each batch run. * Every **10 translated segments**, the file is rewritten in full – so at most 10 segments are lost in a crash. * The file is written atomically (via a temp file + rename) so it is always a valid, complete TMX – never a partial or corrupt file. * At the end of a completed or cancelled run, the file is flushed one final time. #### Where the files are saved ```plaintext C:\Users\\Supervertaler\trados\batch_backups\ ``` Files are named by timestamp and project name, for example: ```plaintext batch_2026-04-10_14-23-01_YAXINCHENG.tmx ``` The exact path is also printed to the **Batch Translate log** at the start of each run. #### Recovering after a crash 1. Reopen Trados Studio and your project. 2. Open your TM in **Trados Translation Memories** (or via the project’s TM settings). 3. Use **Import** → browse to the backup `.tmx` file → import. 4. Run **Pre-translate** on your project to apply the recovered translations from the TM. Note Backup files are **not deleted automatically**. Tidy up the `batch_backups` folder occasionally if disk space is a concern, or keep them as a translation archive. ### Clipboard Mode If you prefer to use a web-based AI (ChatGPT, Claude, Gemini, etc.) instead of an API, tick the **Clipboard Mode** checkbox. This replaces the Provider and Translate button with **Copy to Clipboard** and **Paste from Clipboard** buttons. Supervertaler builds a complete, ready-to-use prompt – including your selected prompt, terminology, document context, and numbered bilingual segments – and copies it to your clipboard. See [Clipboard Mode](/trados/clipboard-mode/) for full details. ### Tips #### Translate Empty Segments First Start by translating only the empty segments (scope: **Empty Segments Only**). Review the results, then fix any issues. This avoids overwriting segments you have already edited. #### Generate a Domain-Specific Prompt Automatically Click **AutoPrompt…** next to the prompt dropdown. Supervertaler analyses your entire document, detects the domain, and uses AI to generate a comprehensive translation prompt with terminology rules, style guidelines, and anti-truncation controls – all tailored to your specific project. See [AutoPrompt](/trados/generate-prompt/) for details. #### Create Domain-Specific Prompts Manually For specialised content, you can also duplicate the Default Translation Prompt in the Prompt Manager and add domain-specific instructions (terminology rules, style preferences, formatting requirements). A tailored prompt is the single most effective way to improve translation quality. #### Combine with TM If your project has a translation memory, TM matches are shown alongside AI translations. You can pre-translate with TM first (using Trados’s built-in batch tasks), then use Batch Translate to fill in the remaining empty segments with AI. #### Review After Batch AI translation is a first draft. After a batch run: 1. Review each translated segment 2. Fix any terminology or style issues 3. Confirm segments with **Ctrl+Enter** (Trados default) *** ### See Also * [Clipboard Mode](/trados/clipboard-mode/) * [AutoPrompt](/trados/generate-prompt/) * [AI Proofreader](/trados/ai-proofreader/) * [Supervertaler](/trados/ai-assistant/) * [TermLens](/trados/termlens/) * [SuperMemory](/trados/ai-assistant/super-memory/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) # Clipboard Mode Clipboard Mode lets you translate or proofread segments using **any web-based AI** – ChatGPT, Claude, Gemini, DeepSeek, or any other LLM with a chat interface – without needing an API key. Instead of sending segments to an AI provider via API, Supervertaler builds a ready-to-use prompt and copies it to your clipboard. You paste it into the AI of your choice, copy the response, and paste it back. Tip **No API key? No problem.** Clipboard Mode is the fastest way to start using AI translation in Supervertaler for Trados. If you already have access to a web-based AI chat – and most people do these days – you can start translating immediately after installing the plugin. No API keys, no provider configuration, no per-token billing. Just tick Clipboard Mode, copy, paste, and translate. Clipboard Mode is also ideal if you want to use a model that is not available via API, if you prefer a pay-as-you-go chat subscription, or if you want to try different AI models before committing to a specific provider’s API. ## How It Works Clipboard Mode is available in both **Translate** and **Proofread** modes on the Batch Operations tab. ### Translating with Clipboard Mode 1. Open the **Supervertaler Assistant** panel and switch to the **Batch Operations** tab 2. Set the mode to **Translate** 3. Tick the **Clipboard Mode** checkbox 4. Choose a **scope** (Empty Segments Only, All Segments, etc.) 5. Optionally select a **prompt** to customise the translation instructions 6. Click **Copy to Clipboard** 7. Open your preferred web-based AI (ChatGPT, Claude, Gemini, etc.) 8. Paste the prompt into the chat and send it 9. Copy the AI’s full response 10. Switch back to Trados and click **Paste from Clipboard** The translations are written into the target segments automatically, with full tag reconstruction and validation. ### Proofreading with Clipboard Mode 1. Set the mode to **Proofread** and tick **Clipboard Mode** 2. Click **Copy to Clipboard** – the prompt includes both source and target text for each segment 3. Paste into your AI, copy the response, and click **Paste from Clipboard** ## What Gets Copied When you click **Copy to Clipboard**, Supervertaler builds a comprehensive prompt that includes: * **System instructions** – the same translation or proofreading instructions used by the API-based batch modes * **Custom prompt** – your selected prompt from the Prompt Manager, if any * **Terminology** – terms from your enabled termbases, including definitions and domains (when term metadata is enabled in AI Settings) * **Document context** – source segments from the document (when enabled in AI Settings), so the AI understands the document type and domain * **Numbered bilingual segments** – each segment is numbered and formatted with status annotations This is not just a list of segments – it is a fully self-contained prompt ready to paste into any LLM chat window. ### Segment Format Each segment is formatted as a numbered bilingual block: ```plaintext Segment 1 [new]: Dutch: Polyvision breidt de mogelijkheden uit. English: Segment 2 [fuzzy, 85%]: Dutch: Nieuwe toepassingen in onderwijs. English: New applications in education. Segment 3 [translated, 100%]: Dutch: Doelstelling lange termijn. English: Long-term objective. ``` The per-segment labels use short language names (e.g. “Dutch”, “English”) to save tokens. The full language names with regional variants (e.g. “Dutch (Netherlands)”, “English (United Kingdom)”) are stated once in the system prompt at the top. ### Status Annotations Each segment header includes a status annotation in square brackets: | Status | Meaning | | ------------------------- | ---------------------------------------- | | **\[new]** | No target text – needs translation | | **\[fuzzy, N%]** | TM fuzzy match at N% – may need revision | | **\[translated, 100%]** | 100% TM match – likely correct | | **\[translated]** | Human-edited translation | | **\[machine translated]** | Machine translation output | | **\[draft]** | Has target text but origin is unclear | These annotations help the AI understand the state of each segment and respond appropriately – for example, a fuzzy match may only need minor adjustments rather than a full retranslation. ## Tag Handling Inline tags (bold, italic, hyperlinks, field codes, etc.) are serialised as numbered placeholders before being sent to the AI: | Tag type | Placeholder | | ---------------- | ---------------------- | | Opening tag | ``, ``, etc. | | Closing tag | ``, ``, etc. | | Self-closing tag | ``, ``, etc. | For example, a segment like “Click **here** for details” becomes: ```plaintext Click here for details ``` The AI is instructed to preserve these placeholders exactly as they appear. When you paste the response back, Supervertaler reconstructs the original Trados tags from the placeholders – the same tag reconstruction pipeline used by API-based Batch Translate. Caution If a tag is missing or malformed in the AI’s response, Supervertaler reports a warning but still writes the translation. Check the log for any tag validation messages. ## Choosing an AI Model Any web-based LLM with a chat interface works with Clipboard Mode. Some recommendations: * **Claude** (claude.ai) – excellent at following the bilingual format precisely and preserving tags * **ChatGPT** (chatgpt.com) – widely available, works well with the structured format * **Gemini** (gemini.google.com) – large context window, good for bigger batches * **DeepSeek** (chat.deepseek.com) – strong multilingual capabilities For best results, use the most capable model available in your subscription (e.g., Claude Opus, GPT-4o, Gemini Pro). Note Most web-based AI chat interfaces have a context limit that determines how many segments you can process at once. If you have a large number of segments, consider using a smaller scope (e.g., Filtered Segments) or processing in multiple rounds. ## Clipboard Mode vs API Mode | | Clipboard Mode | API Mode | | -------------------- | ---------------------------------------------- | -------------------------------------------- | | **API key required** | No | Yes | | **Setup time** | None – works immediately | Requires provider account and API key | | **Cost** | Included in your AI chat subscription | Pay-per-token via API | | **Automation** | Manual copy/paste | Fully automatic | | **Model choice** | Any web-based LLM | OpenAI, Anthropic, Google, Ollama | | **Best for** | Getting started, quick jobs, trying new models | Large projects, automation, batch processing | Both modes use the same prompts, terminology, document context, and tag handling – the only difference is how the text gets to and from the AI. Note Many users start with Clipboard Mode to explore AI translation with zero setup, then move to API Mode later for larger projects where full automation is more efficient. The two modes complement each other – you can switch between them at any time by ticking or unticking the Clipboard Mode checkbox. ## Combining with AutoPrompt – the hybrid pattern [AutoPrompt](/trados/generate-prompt/) **always uses your configured AI provider** to generate the meta-prompt – Clipboard Mode does **not** apply to AutoPrompt, only to the actual Translate / Proofread passes. This is intentional, and it enables a useful hybrid workflow: 1. Tick **Clipboard Mode**. 2. Click **AutoPrompt…** – the prompt-generation request still goes through your configured API (a small, one-shot call, even on an Opus-class model this is cheap relative to bulk translation). 3. Refine the generated prompt in the AI Assistant chat and **Save as Prompt…**. 4. Select the saved prompt from the dropdown. 5. Click **Copy to Clipboard** – Supervertaler builds a ready-to-paste batch for your web-based AI using the AutoPrompt-generated system prompt. 6. Paste into ChatGPT / Claude.ai / Gemini, copy the response, click **Paste from Clipboard**. The result: paid API for the small-but-clever prompt-writing call, free web tier for the expensive bulk translation. You get the full AutoPrompt analysis pipeline (TermScan, domain detection, sampling of confirmed reference pairs) without paying per-token API rates for the bulk Translate. Note If you want to run AutoPrompt without using any API key – for example before you have signed up for a provider – you’ll need to use API Mode-only features (the in-app Chat panel) at least once for the prompt-generation step. There is currently no clipboard-only AutoPrompt path; AutoPrompt always sends the meta-prompt request to a configured provider. ## Tips ### Use the Best Model Available Since you are not paying per token in Clipboard Mode, there is no cost difference between models. Use the most capable model your subscription offers. ### Check the Response Format Before clicking Paste from Clipboard, glance at the AI’s response to make sure it followed the numbered bilingual format. Most modern LLMs handle this correctly, but if the format is off, you can ask the AI to reformat its response. ### Combine with Terminology Clipboard Mode includes your termbase terms in the prompt, just like API mode. Make sure your termbases are set up and enabled in AI Settings for the best results. ### Name Prompts After Your Projects If you save a custom prompt with the same name as your Trados project (e.g. “HAYNESPRO” for a project called HAYNESPRO), the prompt dropdown will auto-select it whenever you open that project. This works for both Translate and Proofread modes and saves you from having to reselect the correct prompt each time. ### Process in Batches For large documents, use the scope dropdown to work through segments in manageable batches – for example, use display filters to select a section at a time, then use the **Filtered Segments** scope. *** ## See Also * [Batch Translate](/trados/batch-translate/) * [AI Proofreader](/trados/ai-proofreader/) * [Batch Operations](/trados/batch-operations/) * [AI Settings](/trados/settings/ai-settings/) * [Prompts](/trados/settings/prompts/) # Context layers > The ten layers of context Supervertaler for Trados puts in front of the AI, what each one adds, and how to control them. A translation engine that sees only the sentence in front of it will translate that sentence well and the document badly. Supervertaler’s design is the opposite: every time you send a chat message, translate a batch of segments, or ask AutoPrompt to draft a prompt, it assembles a fresh snapshot of your work and hands the whole thing to the AI. That snapshot is built in **layers**. Some come from Trados, some from your own knowledge, and two of them come from parts of the document that no CAT tool normally shows you at all. They stack, and each one that is present removes a class of mistake the AI would otherwise make. This page is the single place that lists every layer. Each section is a short overview with a link to the feature’s own page. ## The ten layers | # | Layer | Where it comes from | On by default | | -- | ---------------------------- | --------------------------------------- | ------------------------------ | | 1 | Project and file information | Trados | always | | 2 | Current segment | Trados | always | | 3 | Surrounding segments | Trados | always | | 4 | Full document content | Trados | yes | | 5 | A translation memory match | your TMs | yes | | 6 | Termbase terms | your termbases | yes | | 7 | SuperMemory | what you have recorded about the client | yes, when a bank is active | | 8 | List numbering | the Word file inside the sdlxliff | yes, from v18.20.189 | | 9 | Figure descriptions | the images in your documents | after two clicks in FigureLens | | 10 | Attached files | you, per chat turn | when you attach something | Layers 1 to 8 need no work from you at all (layer 5 only where Studio has already put a TM hit in the segment). Layer 9 is two clicks per project. Layer 10 is deliberate. ### 1. Project and file information Which project and file you are in, the language pair (e.g. Dutch → English), and your position in the document (“Segment 42 of 318”). Always included; no toggle. ### 2. Current segment The source text you are translating and any target you have already entered. Always included – the minimum context for most AI operations. ### 3. Surrounding segments Two segments before and two after, with their translations where available. This is what lets the AI resolve a pronoun the way the previous sentence resolved it, or continue a clause that began in the segment above. Always included; the window is fixed. ### 4. Full document content All source segments in the current document, so the assistant can judge what kind of document it is – legal, medical, technical, marketing, financial, scientific – and let that inform its terminology and register. Very long documents are truncated to a configured maximum (default 500 segments), keeping the first 80 % and the last 20 % so the beginning and the end both survive. **Toggle:** AI Settings → *Include full document content*. ### 5. A translation memory match Where Studio has already put a TM hit in the segment – a pre-translated or auto-propagated row – that translation goes to the AI as work you have already approved, so it is followed rather than quietly rewritten. Two limits worth knowing, because they are easy to assume away: * It is the match **Studio left on the segment**, not a search of your TMs for the best few. A segment with no TM origin contributes nothing here, so pre-translating before a batch run is what gives this layer anything to say. * **Batch Translate sends exact matches only.** Studio records how close a match is but not the source it was made for, so a fuzzy could only be offered as a translation of a sentence the model cannot read – unable to tell which words differ, and most misleading precisely where a memory has been padded with near-misses to manufacture matches. At 100% the match’s source *is* the segment’s source, so nothing is hidden. Chat, QuickLauncher and AutoPrompt still send whatever match is on the segment, with its percentage. **Toggles:** **Send 100% TM matches to the AI** on the Batch Operations tab, for a job whose memory you do not trust; AI Settings → *Include TM matches* for the rest. ### 6. Termbase terms Matched terms from your active termbases, with approved translations and synonyms, and optionally definitions, domains and usage notes. Non-translatable and forbidden terms are flagged so the AI respects them. Both Supervertaler termbases and MultiTerm `.sdltb` termbases attached to the project contribute. See [TermLens](/trados/termlens/) and [MultiTerm Support](/trados/multiterm-support/). **Toggles:** AI Settings → *Include termbase terms* / *Include term metadata* / the per-termbase contribution list. ### 7. SuperMemory [**SuperMemory**](/trados/ai-assistant/super-memory/) is where you record the things about a client that cannot be looked up. If a bank is active, its files go out with every AI call: * `brief.md` – who the client is and anything standing * `terminology.md` – term decisions, one table * `style.md` – prose rules and approved boilerplate The `_shared` bank is sent alongside as house defaults, and the active bank is marked as overriding it where the two disagree. Where a termbase gives the AI flat pairs of terms, SuperMemory gives it the **reasoning** behind them: the decisions, the caveats, the client-specific overrides. Only the active bank is used (plus `_shared`), so switch to the right one before translating. **Toggles:** AI Settings → *Include memory bank context* / *Use memory bank in AutoPrompt*. ### 8. List numbering Word numbers claims, letters steps and bullets lists as paragraph properties, not as text – so the segment grid never contains the `a)` or the `9.`, and neither did anything the AI received. On a real patent that produced a model reading six unlettered steps, translating “steps a. to f.” faithfully, and then flagging it as a possible defect in the source: a note that would have reached the client. Batch Translate, Translate Segment, Clipboard Mode and SuperBench now prefix the first segment of each numbered paragraph with the marker Word renders, inside a sentinel – `[#e)]`, `[#9.]`, `[#•]` – and a rule in Supervertaler’s own preamble tells the model it is structure, to be used for cross-references and parallelism and never reproduced. Anything the model echoes back is stripped before the target is written. The markers are read from the original Word file Studio keeps inside the sdlxliff and computed for the whole document at once, so a list restarting at claim 11 reads `11.` and lettered steps that continue across claims keep counting, exactly as Word shows them. Always on from v18.20.189; see [Batch Operations](/trados/batch-operations/#list-numbering-as-structure-context). ### 9. Figure descriptions The AI reads your text and cannot see your pictures. [**FigureLens**](/trados/batch-operations/#figurelens-from-v1820189) closes that gap: it takes the images out of the project’s Word documents into a folder, shows each one to the AI together with what the document says about it, and saves a description – what it shows, which figure it is, which parts and reference numbers appear on it – as `figures.md` in the active memory bank. From then on it rides along with layer 7 and is read with every request, so the model knows that the “valve (12)” in the sentence is the thing at the top right of Figure 3. Vector drawings (EMF, WMF – what a drawing placed from CAD usually is) are rendered to PNG on the way out, because no AI can read a metafile. Two clicks per project, then it is automatic. One AI request per image; the button says how many and to which provider before you click, and **Describe from the text only** is a free alternative that uses what the document itself says about each figure. ### 10. Attached files Files you attach to a chat turn – images by paste, drag-drop or browse, and documents (DOCX, PDF, PPTX, XLSX, CSV, TMX, SDLXLIFF, TBX, TXT, Markdown, HTML and more). Images go through each provider’s vision API; documents are text-extracted and appended. Attachments apply only to the turn you attached them on. See [File Attachments](/trados/ai-assistant/file-attachments/). ## Stacking the layers All ten combine freely, and the default composition – everything above except the two that need a click – is a strong baseline and the one to start from. More context is not automatically better, though. The context window is finite, and a large project with a rich termbase and a mature memory bank can push a prompt into the 50 000–100 000-token range. At some point: * adding **TM matches** on top of a memory bank that already knows the client’s preferred wordings may add noise rather than signal; * including **full document content** for a very long document may leave too little room for the memory bank to load; * layering **three overlapping sources** (TM + termbase + memory bank) on the same concept may produce contradictions the AI has to reconcile on the fly. Note **Composing tip:** for a client where you have a well-built memory bank, try a small batch with TM matches disabled and compare it to a run with everything enabled. The cleaner prompt often gives more consistent terminology, because the bank already knows which wording the client prefers and the TM matches add nothing it does not say better. For unfamiliar domains or one-off jobs, keep everything on – the TM and termbase are carrying the weight there. The four layers that describe the *document* rather than the *client* – 3, 4, 8 and 9 – behave differently: they are cheap, they never contradict each other, and there is no case yet found where turning one off improved a translation. Caution We have not published composition presets (e.g. “Mature client – memory bank only”, “Unfamiliar domain – TM + termbase”). Until we do, leave everything enabled and experiment per project. If you find a configuration that works well, say so in [GitHub Discussions](https://github.com/orgs/Supervertaler/discussions). ## Seeing and controlling the layers To see exactly what is being sent, click **👁 Preview prompt** next to the Translate button on the Batch Operations tab. It opens a read-only dialog with the assembled prompt – every layer, in the order the model receives it – and makes no AI call. To change what is sent, open the settings dialog’s **AI Settings** tab, where you can toggle document content, TM matches, term metadata and memory bank context, and choose which termbases contribute. ## See Also * [Supervertaler Assistant](/trados/ai-assistant/) – overview * [Batch Operations](/trados/batch-operations/) – Preview prompt, FigureLens, list numbering * [AI Settings](/trados/settings/ai-settings/) – the toggles * [SuperMemory](/trados/ai-assistant/super-memory/) – the client decisions you record by hand * [SuperMemory → AI Integration](/trados/ai-assistant/super-memory/ai-integration/) – the loading algorithm and token budget * [File Attachments](/trados/ai-assistant/file-attachments/) – add images and documents to a chat turn * [TermLens](/trados/termlens/) – how termbase terms are matched and loaded # User Data Folder Supervertaler for Trados shares a user data folder with [Supervertaler Workbench](https://supervertaler.com/workbench). This allows both programs to access the same termbases, translation memories, and prompt library without duplicating files. ## Folder Location By default, the shared data folder is located at: ```plaintext C:\Users\\Supervertaler\ ``` You can choose a different location during first-run setup. Both programs read the configured path from the same pointer file at `%APPDATA%\Supervertaler\config.json`. ## Folder Structure ```plaintext Supervertaler/ │ ├── prompt_library/ Shared │ ├── domain_expertise/ │ ├── project_prompts/ │ └── style_guides/ │ ├── resources/ Shared │ ├── supervertaler.db │ ├── termbases/ │ ├── tms/ │ ├── non_translatables/ │ └── segmentation_rules/ │ ├── workbench/ Supervertaler Workbench only │ ├── settings/ │ │ ├── settings.json │ │ ├── themes.json │ │ ├── shortcuts.json │ │ └── ... │ ├── dictionaries/ │ ├── projects/ │ ├── ai_assistant/ │ ├── voice_scripts/ │ └── web_cache/ │ └── trados/ Supervertaler for Trados only ├── settings/ │ ├── settings.json │ ├── license.json │ └── chat_history.json ├── projects/ └── batch_backups/ ``` ### Shared resources The **prompt library** and **resources** folders are shared between both programs. Prompts you create or edit in one program are immediately available in the other. The SQLite database (`supervertaler.db`) holds your termbases and translation memories – Workbench has full read-write access, while the Trados plugin reads from it. ### Program-specific folders Each program stores its own settings, projects, and runtime data in a dedicated subfolder (`workbench/` or `trados/`). This keeps configuration separate so the two programs never interfere with each other. The `trados/batch_backups/` folder contains automatic TMX backup files created during Batch Translate runs. One file is written per run, named by timestamp and project name. These files are not deleted automatically – you can remove old ones manually once your project is safely delivered, or keep them as a translation archive for use in other CAT tools. See [Batch Translate – Backup TMX](/trados/batch-translate/#backup-tmx) for details. ## Automatic Migration If you are updating from an older version, both programs will automatically reorganise the folder on their next startup. No manual action is required – your settings, licence, and data are preserved. # AutoPrompt AutoPrompt uses AI to analyse your entire project and generate a comprehensive, domain-specific translation prompt tailored to your document. The generated prompt includes terminology rules, style guidelines, anti-truncation controls, and domain-specific instructions – ready to use with Batch Translate. ![]() #### How It Works **1. Start the analysis** On the **Batch Operations** tab, click the **AutoPrompt…** link next to the prompt dropdown. **2. What gets analysed** Supervertaler gathers the following data from your project and sends it to your configured AI provider: | Data | Purpose | | ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | **All source segments** | Domain detection, document content analysis, project context | | **Termbase terms** | Filtered to only document-relevant terms (TermScan), then included as a locked termbase in the generated prompt | | **Translated segments** | Human-confirmed segments only (Translated, Approved, or Signed-off status) – used as TM reference pairs and style anchors. Unconfirmed AI-generated translations are excluded. | | **Language pair** | Embedded in the generated prompt | Note The full document is sent to the AI for analysis. For a typical 30,000-word document, this costs approximately $0.20–$0.25 with a Sonnet-class model, or $1.00–$1.15 with an Opus-class model. **2b. TermScan – automatic termbase extraction** Before building the prompt, AutoPrompt runs **TermScan**: it concatenates all source segments in the document and checks each termbase entry against this text. Only terms whose source term, source abbreviation, or source synonyms actually appear in the document are included in the generated prompt. This dramatically reduces the termbase size – for example, a general patent termbase with 2,680 entries might yield only 123 relevant terms for a specific document. The status message in the AI Assistant confirms the filter: *“Termbase terms (filtered 123 relevant from 2,680 total)”*. The filtering is case-insensitive and checks all variants of each term (source term, abbreviation forms, and synonyms). Terms that do not appear anywhere in the source text are excluded entirely. Caution **TermScan filters by content, not by domain.** A patent termbase contains many common technical words – “system”, “board”, “fan”, “screen”, “installation” – that will match nearly any engineering document. TermScan will include those entries even though they belong to a completely different translation context, and the AI will be forced to follow them. **Before running AutoPrompt, disable every termbase that does not belong to the current project’s domain.** Use [AI Settings](/trados/settings/ai-settings/) → *Termbases included in AI prompts* to control which termbases contribute to the generated prompt without affecting your TermLens display. Caution **Termbase quality matters.** Only enable termbases in [AI Settings](/trados/settings/ai-settings/) if you are confident they contain accurate, high-quality terminology for your project. A poorly maintained termbase with incorrect or outdated translations will constrain the AI and produce worse results. Modern LLMs – especially Opus-class and GPT-4-class models – are often better at choosing the right translation on their own than when forced to follow a low-quality termbase. When in doubt, disable termbases and let the AI translate freely, then add terms incrementally as you review. **3. Domain detection** Before sending to the AI, AutoPrompt runs a local keyword-based analysis to detect the document’s domain. Supported domains: * **Patent** – claims, embodiments, prior art, figure references * **Legal** – contracts, clauses, statutory references * **Medical** – clinical terms, dosages, ICD/ATC codes * **Technical** – specifications, software terms, standards * **Financial** – figures, IFRS/GAAP, regulatory language * **Marketing** – brand, audience, campaign language * **General** – fallback for mixed or unclassified content The detected domain determines which template the AI uses to generate the prompt – including domain-specific roles, rules, and section structure. **4. Review and refine in the AI Assistant** The generated prompt appears as a message in the **AI Assistant** chat. You can: * **Read through the prompt** to verify it matches your project * **Ask follow-up questions** to refine specific sections (e.g., “Make the termbase section more strict” or “Add a rule about chemical formula formatting”) * **Iterate** as many times as needed – each refinement builds on the conversation history **5. Save the prompt** When you are satisfied with the generated prompt: 1. **Right-click** the assistant message containing the prompt 2. Select **Save as Prompt…** 3. Enter a name for your prompt (e.g., “DLCH Patent NL-EN”) 4. Click **Save** The prompt is saved to the **Translate** category in the Prompt Manager and immediately appears in the prompt dropdown on the Batch Operations tab. Note **Generated prompts are formatted in proper Markdown** – `##` headings for each major section, `-` bullet lists, `**bold**` for emphasised terms, and a Markdown table for the project-specific termbase. Open one in Obsidian, VS Code, GitHub, or any Markdown-aware viewer and it renders cleanly with a navigable outline. The prompt is also still a perfectly valid system prompt for the translator AI – the Markdown markup is structural, not output instruction. #### Translator’s Comment methodology (always-on) Since v4.19.111, every AutoPrompt-generated prompt embeds the **Translator’s Comment** (TC) methodology by default, regardless of source language or domain. The methodology asks the translator AI to silently correct obvious mechanical defects in the source (typos, broken words, hanging mid-sentence breaks, doubled spaces, stray punctuation, reference-numeral mismatches that are unambiguous in context, missing diacritics, etc.) and append a single concise comment at the end of the segment in this exact format: ```plaintext ⟦TC: short factual description of the fix(es)⟧ ``` * The brackets are the mathematical white square brackets **U+27E6** (⟦) and **U+27E7** (⟧). These characters do not occur in source documents, so they are safe as out-of-band markers that can be extracted reliably in post-processing. * One marker per segment maximum; multiple fixes are joined with semicolons inside one marker. * Segments with no defects emit no marker. * When the translator AI inserts a word or short phrase to fill a clear gap, that supplied text is wrapped in standard ASCII square brackets `[like this]` inside the running translation, and the trailing marker references it (e.g. `⟦TC: [bracketed text] supplied to close hanging sentence⟧`). * Numerical values, dates, dosages, claim language, statutory references, headings, identifiers, and proper names are never silently “corrected” – defects in those zones are preserved verbatim, with an optional `⟦TC: source ambiguous – ...⟧` marker if doubt exists. The defect categories that count as “obvious” are adapted to the actual source language by the LLM (Dutch -d/-t verb typos, German missing umlauts, French accent slips, Spanish/Italian conjugation typos, etc.). Note **The markers appear inline in the target segment** in Trados Studio – they are not yet auto-extracted into the Studio comments panel. Extraction into the existing Studio comment infrastructure is a planned follow-up; the spec is locked (⟦ and ⟧ delimiters never collide with source text) so the extraction step is small once it gets prioritised. Caution **Want a generated prompt without the TC methodology?** Edit the generated prompt in the Prompt Manager after creation and remove the TRANSLATOR COMMENT FORMAT section plus any TRANSLATION MANDATE language about silent correction. A per-project opt-out via a UI toggle may be added in a future version – open an issue if you’d like to see it. #### What the Generated Prompt Contains A generated prompt follows the structure of professional translation prompts used by experienced translators. Depending on the domain, it typically includes: * **Role** – domain-specific translator role with expertise areas * **Translation mandate** – strict rules against simplification, paraphrasing, or “improving” the source * **Anti-truncation controls** – explicit prohibition of omitting repetitive phrases or collapsing clauses * **Input handling rules** – instructions for segment-by-segment translation in Supervertaler * **Domain-specific style rules** – mandatory term mappings, register requirements, formatting rules * **Terminology hierarchy** – priority order: TM matches > project termbase > domain conventions * **Preflight self-check** – internal verification step before producing output * **Post-translation integrity assertion** – completeness and faithfulness check * **Project context** – AI-generated summary of what the document is about * **Project-specific termbase** – all termbase terms, marked as locked and mandatory * **TM reference translations** – validated translation pairs as style anchors * **Output format** – translation only, no commentary, preserve formatting #### Tips **Start with a confirmed translated sample** The generator includes **confirmed** segments (Translated, Approved, or Signed-off status) as reference pairs – up to 50, sampled evenly across the document. This gives the AI concrete examples of your preferred style and terminology, resulting in a more accurate prompt. Unconfirmed segments (e.g. from a previous AI batch translation that you haven’t reviewed yet) are excluded to avoid feeding unverified output back as “correct” references. **Tip:** Before generating a prompt, confirm a handful of segments you are happy with. Even 10–20 confirmed segments give the AI meaningful style anchors to work from. **Only enable termbases that belong to this project** Before clicking AutoPrompt, go to **AI Settings → Termbases included in AI prompts** and enable only termbases that are directly relevant to the current project. Disable everything else – including large general-purpose termbases, termbases from other clients or domains, and any termbase you are not actively maintaining for this project. This setting is independent of your TermLens display: disabling a termbase for AI context does not hide its chips in the editor. You can keep a termbase visible for reference while excluding it from the generated prompt. Note **Prefer compact, project-specific termbases.** A small termbase of 50–200 carefully curated entries for this client will produce a far better termbase than a general termbase of 2,000 entries, even after TermScan filtering. Large termbases increase the chance of incorrect or misleading entries being injected into the prompt. **Review the termbase section** The generated prompt includes only the document-relevant terms extracted by TermScan from your enabled termbases. Check that the termbase accurately reflects your terminology preferences. You can ask the AI to reorganise terms by category or add missing mappings. Caution If your termbase contains incorrect or low-quality entries, these will be injected into the prompt and the AI will be forced to follow them. Only enable termbases that you trust. When starting a new project with no established terminology, consider disabling termbases entirely and letting the AI translate freely – then add terms as you review. **Use with Batch Translate** After saving the generated prompt, select it from the prompt dropdown on the Batch Operations tab. It works with all scopes and providers, just like any other prompt. **AutoPrompt always uses your configured AI provider** [Clipboard Mode](/trados/clipboard-mode/) does **not** apply to AutoPrompt – ticking the Clipboard Mode checkbox affects only the actual Translate / Proofread passes, not prompt generation. AutoPrompt always sends the meta-prompt request to whichever provider is selected in [AI Settings](/trados/settings/ai-settings/). This enables a useful pattern: keep Clipboard Mode ticked, click AutoPrompt to generate the prompt via your paid API, then run the bulk Translate via clipboard against a free web-tier model. See [Combining with AutoPrompt – the hybrid pattern](/trados/clipboard-mode/#combining-with-autoprompt--the-hybrid-pattern) for the full workflow. **Regenerate when the project changes** If your project evolves significantly (new terminology, different document sections, additional termbases), run AutoPrompt again to generate an updated prompt. *** #### See Also * [Batch Translate](/trados/batch-translate/) * [Prompts](/trados/settings/prompts/) * [Supervertaler](/trados/ai-assistant/) * [AI Settings](/trados/settings/ai-settings/) # Getting Started This page walks you through the first-time setup so you can start using TermLens terminology and AI translation inside Trados Studio. Tip **Prefer to watch?** The [Getting Started screencast](https://www.youtube.com/watch?v=bOIwMAoP7xc) (16 min) covers everything on this page and more – TermLens, prompt generation, AI translation, the Chat window, and purchasing. ## First-Time Setup ### 1. Open Settings Click the **gear icon** in the TermLens panel header to open the settings dialogue. ### 2. Configure Termbases (TermLens tab) On the **TermLens** tab: 1. Click **Browse** to select an existing Supervertaler termbase (`.db` file) 2. Or click **New** to create a new empty termbase You can add multiple termbases. Designate one as the **Project termbase** to give its terms higher priority (shown in pink). Note Supervertaler for Trados uses the same `.db` termbase format as Supervertaler Workbench. Any termbase created in either tool works in both. On Windows, both tools can point to the same `.db` file in a shared data folder. On a Mac running Trados via Parallels, the two products use separate filesystems – see [Running on a Mac](/trados/installation/#running-on-a-mac-parallels) for details. ### 3. Configure AI (AI Settings tab) On the **AI Settings** tab: 1. Select a **provider** (OpenAI, Anthropic, Google, OpenRouter, Ollama, or others) 2. Enter your **API key** for the selected provider 3. Choose a **model** Note **Don’t have an API key yet?** You can skip this step entirely and use **[Clipboard Mode](/trados/clipboard-mode/)** instead. Clipboard Mode lets you translate and proofread using any web-based AI you already have access to – ChatGPT, Claude, Gemini, or any other LLM chat interface. No API key required. It is the fastest way to start using AI translation in Supervertaler for Trados. ### 4. Click OK Settings are saved and applied immediately. ## Try It Out ### TermLens 1. Open a project in the Trados **Editor** view 2. Navigate to any segment – TermLens automatically displays term matches for the source text 3. Click a term translation to insert it into the target, or press **Alt+1** through **Alt+9** ### Clipboard Mode (no API key needed) 1. Open the **Supervertaler Assistant** panel and switch to the **Batch Operations** tab 2. Tick the **Clipboard Mode** checkbox 3. Click **Copy to Clipboard** – a ready-to-use prompt with your segments, terminology, and instructions is copied 4. Paste it into any web-based AI (ChatGPT, Claude, Gemini, etc.) and send it 5. Copy the AI’s response and click **Paste from Clipboard** – the translations are written back into Trados See [Clipboard Mode](/trados/clipboard-mode/) for the full walkthrough. ### AI Translate (API key required) 1. Place the cursor in a segment 2. Press **Ctrl+T** to translate the active segment with AI 3. The AI translation appears in the target cell ### Supervertaler 1. Open the **Supervertaler Assistant** panel (View > Supervertaler Assistant) 2. Switch to the **Chat** tab 3. Type a question about the current segment and press **Enter** 4. The assistant responds with context from your terminology and TM matches ## Quick Links | Feature | Page | | ----------------------------- | ------------------------------------------------- | | TermLens terminology display | [TermLens](/trados/termlens/) | | AI via clipboard (no API key) | [Clipboard Mode](/trados/clipboard-mode/) | | AI chat interface | [Supervertaler](/trados/ai-assistant/) | | Bulk AI translation | [Batch Translate](/trados/batch-translate/) | | All shortcuts | [Keyboard Shortcuts](/trados/keyboard-shortcuts/) | *** ## See Also * [Installation](/trados/installation/) * [Clipboard Mode](/trados/clipboard-mode/) * [TermLens](/trados/termlens/) * [Supervertaler](/trados/ai-assistant/) # Import/Export The **Import/Export** tab in the Supervertaler Assistant panel exports the active Trados document’s segments to a proofreader-friendly file (Word DOCX, Bilingual Text, or HTML), then re-imports the proofreader’s edits back into Trados with a confirmation diff.  This is the workflow you’d use for: * **External review** – send a bilingual DOCX to a colleague who doesn’t have Trados, get it back with edits, apply. * **Quick AI proofreading via web LLM** – copy the bilingual text into ChatGPT/Claude/Gemini, paste the corrected version back, re-import. * **Multi-file project review** – export every file in a merged project into one combined DOCX with section breaks between each source file. ## Formats The export offers three formats, matching the Supervertaler Workbench: | Format | Re-importable | When to use | | --------------------------------------- | ------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Word document (.docx)** | ✅ | The default. A 5-column bilingual table (`#`, source, target, status, notes) the proofreader edits in Word. Identical to the Workbench’s [Bilingual Table](/workbench/import-export/bilingual-tables/), so files move between both products. | | **Bilingual Text (AI-friendly) (.txt)** | ✅ | A compact plain-text format – one block per segment – ideal for pasting into ChatGPT / Claude / Gemini or editing in any text editor. Identical to the Workbench’s [Bilingual Text](/workbench/import-export/bilingual-text/). | | **HTML report (.html)** | ❌ | Client-facing read-only report. Cannot be re-imported. | There’s no separate “layout” picker: each format has one shape – DOCX and HTML use the 5-column table, and Bilingual Text uses the bracketed `[SEGMENT NNNN]` blocks below. Note **Why “Text” and not “Markdown”?** The `.txt` file is deliberately plain text: its segment blocks rely on line breaks being preserved, which a Markdown renderer would collapse. AI agents read the raw characters when you paste the file into a chat, so plain text is both safe and maximally readable. *(Earlier versions offered a Markdown (.md) format and stacked source/target layouts; these were retired in favour of the two round-trippable formats above. Files exported by those older versions can still be re-imported.)* ## The Bilingual Text format Each segment is one block, blank-line separated, with 2-letter language codes labelling the source and target lines: ```plaintext [SEGMENT 0001] EN: MASHUP APPLICATION PROCESSING SYSTEM NL: MASHUP-APPLICATIEVERWERKINGSSYSTEEM [SEGMENT 0002] EN: FIELD OF THE INVENTION NL: GEBIED VAN DE UITVINDING ``` * The `EN:` line is the **source** – leave it alone. It stays on **one line**; a `[newline]` token in it marks where the original source broke across two lines (read-only reference, never written back). * The `NL:` line is the **target** – edit it freely, but **keep it on one line**. Where the target needs a hard line break (e.g. to split a subtitle across two lines), write the literal token `[newline]`; on re-import it’s turned back into a real break. Older files that wrapped a field over several physical lines still re-import unchanged. * The `[SEGMENT NNNN]` markers are the alignment anchors – don’t rename them. Because each field is one labelled line (not a table column), it survives pipe characters and long inputs without the source/target roles getting confused, and keeping targets to one line stops an LLM from accidentally reflowing them. This matches the Workbench’s Bilingual Text export byte-for-byte, so a file produced by either tool round-trips through the other. ## Inline formatting markers Source and target cells use semantic placeholders for inline formatting so the proofreader can move them around without breaking Trados: * `...` – bold pair * `...` – italic pair * `...` – underline pair * `...` – bold + italic pair * `...`, ``, … – numbered placeholders for everything else (field codes, page numbers, custom format pairs) In the DOCX, the markers render in red and the text between a semantic pair is shown in matching bold / italic / underline so the proofreader can see what the formatting will look like. **Round-trip rules:** * Semantic markers (`` / `` / `` / ``) can be freely **added, removed, or reordered** in the target – they only affect cosmetic rendering and don’t drive Trados QA. * Numbered structural markers (``, ``, …) **must round-trip exactly**. Adding a `` the source doesn’t have, or dropping one the source requires, would break the Trados file. The importer counts numbered markers on both sides and skips any segment whose count has changed (with a per-segment log entry). The **“Refuse to apply edits that drop source-required tags”** checkbox enables this strict check. Leave it on unless you know what you’re doing. ## Multi-file projects When the active Trados editor view contains more than one file merged (common when the Trados project was prepared with file merging), the tab grows extra controls: ### Files to export A checkbox list of every file in the active document, each with its segment count. Quick-select buttons: * **Active only** – checks only the file your cursor is currently in * **All** – checks every file * **None** – unchecks every file (Segments: 0) The **Segments: N** label tracks the current selection live. ### Output mode * **Combine into one file** (default) – produces a single bilingual file containing all selected files joined together, with a file-boundary marker between each source file so the proofreader can see where one file ends and the next begins. In a **DOCX** the table grows a 6th **File** column with a highlighted ”📄 File: ``” section-break row; in a **Bilingual Text** file a `📄 File: ` marker line prefaces each new file’s first segment. * **Separate file per file** – asks for a folder and writes one bilingual file per selected source file. Single-file documents see no change – the file list, output radio, and per-file UI are all hidden. (On re-import, the DOCX table’s 5- vs 6-column form is auto-detected.) ## Locked segments Trados segments can be **locked** (read-only in the editor) independently of their confirmation level – so a segment can be both `ApprovedTranslation` and locked, or `Draft` and locked. On large projects with a lot of locked-approved content, sending those segments to a proofreader is usually noise: any edits they make there can’t be written back to Trados anyway. A dedicated checkbox controls how the export handles them: > **Include locked segments (🔒 marked in Status column)** – default ON. * **ON (default)** – locked segments are exported alongside everything else, and every locked row gets a **🔒** prefix in the Status column (e.g. `🔒 ApprovedTranslation`). The proofreader can see at a glance which rows aren’t editable round-trippable, and the re-import will refuse to overwrite them. * **OFF** – locked segments are skipped entirely. The exported file only contains rows that are actually still editable. Useful on multi-thousand-segment projects where the bulk of the work is already locked. The checkbox lives right under the **“Refuse to apply edits that drop source-required tags”** option on the tab. **On re-import**, locked segments are honoured regardless of the export-side setting: edits made to a locked row are reported as a *locked-segment* item in the re-import summary’s “other issues” count, and **not** written back to Trados. To genuinely change a locked segment, unlock it in Trados first, then re-import. The locked flag also lives in the sidecar manifest (`is_locked: true` / `false` per segment) so the source of truth for which segments were locked at export time is preserved. ## Filtering by confirmation status Just below the locked-segments option is a **Statuses to include in export** group of six checkboxes – one per Trados confirmation level: * **Unspecified** – not yet translated (the initial state of a fresh segment) * **Draft** – in progress * **Translated** – confirmed by the translator * **Approved (translation)** – first-pass approval * **Approved (sign-off)** – final approval * **Rejected** – marked for rework All six are checked by default – no filter, every segment is included. Untick any subset to narrow the export to just those statuses. Common use-cases: * **Tick only Translated** – send a draft pass to a proofreader. * **Tick only Approved (translation)** – send near-final material out for a sign-off review. * **Untick Approved (sign-off)** – exclude locked-down rows the client has already signed off on, so the proofreader only sees what’s still in play. * **Tick only Draft + Unspecified** – generate a worklist of what’s still unfinished. This filter composes orthogonally with the **Include locked segments** option and the multi-file **Files to export** list – every segment must pass all three filters to make it into the bilingual file. ## Re-import workflow Click **📥 Re-import…**, pick the round-tripped file. Supervertaler: 1. Loads the file’s sidecar manifest (the `.svexport.json` written alongside the export). 2. For each row in the file, looks up the matching segment in Trados via the manifest’s `(ParagraphUnitId, SegmentId)` mapping. 3. Compares the file’s target text against the current Trados target – same serialisation pipeline on both sides, so only real edits register as changes. 4. Counts up: **changes to apply**, **unchanged**, **tag-mismatch** (will be skipped under strict mode), and **other issues** (segment missing, **locked**, source text was tampered with). 5. Shows you a summary dialog with **OK** / **Cancel**. Click **OK** and Supervertaler writes the accepted changes back via the same code path the batch AI translator uses – confirmation level is preserved, and **locked segments are skipped automatically** (the writeback queries `IsLocked` on every segment, so a lock toggled in Trados between export and re-import is also respected). ## Recent exports At the bottom of the tab, a list tracks every export from this session with: * **Open file** – opens the bilingual file in its default app * **Open folder** – opens the containing folder * **Re-import this** – same as the main Re-import button, pre-pointed at this file ## Sidecar manifest Every export writes a small `.svexport.json` file alongside the bilingual file. It contains: * Project name, source filename, language pair, export timestamp, tool version * Per-segment `(number → ParagraphUnitId / SegmentId)` mapping * A SHA-256 prefix of the source text for tamper detection * For multi-file exports: per-segment source file id + name * Per-segment `is_locked: true | false` flag (snapshot at export time) The manifest is what lets re-import find the exact Trados segments even if the proofreader accidentally reorders rows. If the manifest goes missing, re-import falls back to current-document mapping (which loses source-tamper protection but still works). ## See Also * [Batch Operations](/trados/batch-operations/) – for AI-driven proofreading directly in Trados * [AI Proofreader](/trados/ai-proofreader/) – in-Trados proofreading mode # Installation ### Installation #### Download **Install Supervertaler for Trados from the [RWS App Store](https://appstore.rws.com/plugin/432).** This is the recommended route for everyone: every published build is RWS-signed, so Trados loads it without the “Unsigned Trados Studio Plug-in Found” warning that appears – *every time Studio starts* – for a plugin installed from anywhere else. Note Supervertaler for Trados comes in two builds: one for **Trados Studio 2024** and one for **Trados Studio 2026** (which uses the new `.ttb` termbase format). Install the build that matches your Studio version – the 2024 build will not load in Studio 2026, and vice versa. See [Trados Studio 2026 & .ttb](/trados/studio-2026/) for details. The installation steps below apply to both; only the version you select in the Plugin Installer differs. You can either install from inside Trados Studio (**Add-Ins > RWS App Store**, search for “Supervertaler”, click **Download**) or download the `Supervertaler for Trados.sdlplugin` file from the [App Store website](https://appstore.rws.com/plugin/432) and double-click it. Either path opens the Trados Plugin Installer. Note **What about GitHub?** The [GitHub repository](https://github.com/Supervertaler/Supervertaler-for-Trados) holds the source, the issue tracker and the release notes for every build – but **not the plugin itself**. The App Store is the single publication channel, so every install is signed, and the plugin’s built-in update check (which reads the App Store catalogue) can keep you current. App Store updates go through RWS review, so a brand-new fix can take a day or two to appear there. If you are waiting on a specific fix, email and it can be sent to you directly. #### Install 1. **Close Trados Studio** if it is running 2. **Double-click** the downloaded `Supervertaler for Trados.sdlplugin` file 3. The Trados Plugin Installer opens – select your Trados Studio version and choose an installation location:  The Trados Plugin Installer lets you choose which Trados version to install for and where to place the plugin. 4. Click **Next**, then **Finish** to complete the installation 5. **Start Trados Studio** – the plugin loads automatically **Installation locations** The installer offers three options for where to place the plugin. Each option stores the plugin in a different Windows folder, which determines who can use it and whether it follows you to other computers. The paths below are for **Trados Studio 2024**. For **Studio 2026** every path is identical except that `\18\` becomes `\19\`. **“All your domain computers”** (default) : Installs to: `C:\Users\\AppData\Roaming\Trados\Trados Studio\18\Plugins\Packages\` : The Windows **Roaming** profile folder. In environments that sync the Roaming profile across machines – classic Active Directory roaming profiles, FSLogix profile containers, and similar setups – the plugin follows your Windows account from one PC to another. (OneDrive Known Folder Move does **not** sync `AppData\Roaming` by default, so OneDrive on its own is not a roaming mechanism.) On a single-PC personal install without any profile-sync setup, the plugin simply stays on the machine – functionally similar to “This computer for me only”, though the folder is still `Roaming` rather than `Local`, which can matter if the machine later joins a profile-sync environment. **“This computer for me only”** : Installs to: `C:\Users\\AppData\Local\Trados\Trados Studio\18\Plugins\Packages\` : The Windows **Local** profile folder. The plugin stays on this specific machine and is only available to your Windows user account. If another person logs into the same PC with a different Windows account, they will not have the plugin. **“This computer for all users”** : Installs to: `C:\ProgramData\Trados\Trados Studio\18\Plugins\Packages\` : The shared **ProgramData** folder. The plugin is available to every Windows user account on this machine. Use this on shared workstations where multiple people log in with their own Windows accounts and all need the plugin. Rarely needed for most translators. Note **Which should I choose?** **Just leave it on the default** (“All your domain computers”) and click Next. The dialogue always opens with this option pre-selected and it works correctly for everyone. On a single-PC personal install it behaves the same as “This computer for me only” in practical terms (the plugin loads identically) – the only real difference is the install folder (`Roaming` vs `Local`), which only matters in environments that sync the Roaming profile across machines. **All three options work fine** – pick a different one only if you have a specific reason. As long as the **“Remove this plugin from all installation folders”** checkbox stays ticked (it is by default), any orphan-copy issues from previous installs are cleaned up automatically, regardless of which option you pick. **“Remove this plugin from all installation folders” checkbox** The Trados Plugin Installer shows a checkbox below the install-scope radio buttons: > ☑ **Remove this plugin from all installation folders** *(recommended if you installed manually or multiple times)* It is ticked by default. **Always leave it ticked.** Before placing the fresh install in the location you’ve selected, the installer sweeps all three locations (Roaming, Local, ProgramData) and removes any existing Supervertaler copies. On a first-time install there’s nothing to remove and the checkbox is a harmless no-op; on an upgrade or after a previous manual install, it prevents the multi-scope-orphan problem where Trados ends up with two copies of the plugin in different folders and loads the wrong one on next start. #### Verify Installation After restarting Trados Studio, open a project in the Editor view. You should see: * **TermLens panel** – docked above the editor area (or in the bottom panel area) * **Supervertaler Assistant panel** – docked on the right side **If the TermLens panel is not visible** Go to **View > TermLens** to show the panel. **If the Supervertaler Assistant panel is not visible** Go to **View > Supervertaler Assistant** to show the panel. Tip Both panels are standard Trados dockable panels. You can drag them to any docking position (left, right, top, bottom, floating) or move them to a second monitor. Trados remembers their position between sessions. #### Free up the keyboard shortcuts Trados Studio’s own default bindings sit on several of the key combinations Supervertaler uses (`Ctrl+Alt+T`, `Ctrl+Alt+N`, `Ctrl+Alt+G`, `Alt+Up`, `Alt+Q`) – and the Trados binding wins, so those Supervertaler shortcuts do nothing until you clear the defaults. This takes two minutes in **File → Options → Keyboard Shortcuts** and only needs doing once per Trados installation. See [First-time setup: free up Trados shortcuts](/trados/keyboard-shortcuts/#first-time-setup-free-up-trados-shortcuts) for the full table of what to delete. #### Running on a Mac (Parallels) If you are running Trados Studio inside **Parallels Desktop** on a Mac, there is one important rule for the first-run setup: **Keep your data folder on the Windows side** – use the default path (e.g., `C:\Users\\Supervertaler`). Do **not** point it to a Mac-side path like `\\Mac\Home\Supervertaler`. Supervertaler stores termbases as SQLite databases, and SQLite requires a local filesystem to work reliably. The `\\Mac\Home\...` paths in Parallels are mounted via a virtual network share, which can cause database locking errors or data loss. Caution **Mac users:** When the first-run setup dialogue appears, accept the default `C:\Users\\Supervertaler` path. If you previously used Supervertaler Workbench on the Mac side, copy your termbases into the Windows-side folder rather than pointing to the Mac path directly. The plugin automatically detects Parallels and shows a warning if you select a Mac-side path during setup. **Sharing termbases between Workbench and the Trados plugin on a Mac** On Windows, both Supervertaler Workbench and the Trados plugin can point to the same shared data folder and work from the same `.db` termbase file simultaneously. On a Mac with Parallels, this is **not possible** because the two products run on different filesystems: * **Supervertaler Workbench** runs natively on macOS – its data folder is on the Mac filesystem (e.g., `/Users//Supervertaler/`) * **Supervertaler for Trados** runs inside Parallels (Windows) – its data folder must be on the Windows filesystem (e.g., `C:\Users\\Supervertaler\`) The Trados plugin cannot reliably use a Mac-side path (`\\Mac\Home\...`) due to SQLite limitations on virtual network shares. To keep your termbases in sync between the two products on a Mac, copy the `.db` file from one side to the other after making changes. This is a limitation of the Parallels virtualisation layer, not of the termbase format. *** #### Updating When a new version is published to the App Store, Supervertaler shows an **Update Available** dialogue on the next Trados startup, with the version difference and an **Install Update** button. 1. Click **Install Update** – the plugin downloads the RWS-signed update from the App Store and writes it back to the same install scope (Roaming, Local, or ProgramData) you originally chose during installation 2. When prompted, click **Restart Trados Studio** – the plugin restarts Trados for you and loads the new version Your settings, termbases, prompts, memory banks, and licence key are all preserved across updates – no need to uninstall first. **Manual update from the App Store website** If you’ve dismissed the in-plugin dialogue (for example, by clicking **Remind Me Later**) and want to update straight away, you can install manually from the App Store website: 1. **Close Trados Studio completely** – the plugin files are locked while Trados is running 2. Open the [App Store page for Supervertaler](https://appstore.rws.com/plugin/432) and click **Download** to save the latest `Supervertaler for Trados.sdlplugin` 3. Double-click the file – the Trados Plugin Installer handles the rest 4. Start Trados Studio – the new version loads automatically Caution Trados Studio **must be fully closed** before installing or updating manually. If Trados is still running, the installer may silently fail because the plugin files are locked. #### Troubleshooting: old version still showing after update Note From **v4.19.24** onwards the in-plugin updater is install-scope aware – it writes updates back to the same scope (Roaming, Local, or ProgramData) as the original install, so this scenario does not occur for automatic updates. The steps below remain useful if you have inherited a multi-scope install from an earlier version, or have placed `.sdlplugin` files manually in different locations. Tip **Easiest fix:** download the latest `.sdlplugin` from the [App Store page](https://appstore.rws.com/plugin/432), close Trados, double-click the file, and tick the **“Remove this plugin from all installation folders”** checkbox when it appears in the installer. The Trados Plugin Installer will sweep all three install scopes and replace everything with the fresh copy in one step. Manual cleanup steps below are only needed if that path doesn’t work for some reason. If Trados still loads an older version of the plugin after installing a new one, an old copy may be lingering in a different installation location. Check all three plugin folders and remove any old `Supervertaler for Trados.sdlplugin` (in `Packages`) and `Supervertaler.Trados` folder (in `Unpacked`): Replace `` with your own Windows user name. Use the block for your Studio version – `18` is Studio 2024, `19` is Studio 2026. Each path is one of the three choices the Trados Plugin Installer offers under “Please select the folder where the plugin will be installed”, so if you remember which you picked, start with that one – but check all three, because an earlier install may have used a different one. **Trados Studio 2024** ```plaintext C:\Users\\AppData\Roaming\Trados\Trados Studio\18\Plugins\ <- "All your domain computers" (the default) C:\Users\\AppData\Local\Trados\Trados Studio\18\Plugins\ <- "This computer for me only" C:\ProgramData\Trados\Trados Studio\18\Plugins\ <- "This computer for all users" ``` **Trados Studio 2026** ```plaintext C:\Users\\AppData\Roaming\Trados\Trados Studio\19\Plugins\ <- "All your domain computers" (the default) C:\Users\\AppData\Local\Trados\Trados Studio\19\Plugins\ <- "This computer for me only" C:\ProgramData\Trados\Trados Studio\19\Plugins\ <- "This computer for all users" ``` Inside each one there are two folders that matter, and **both** need clearing: * `Packages` holds the installed `.sdlplugin` file. * `Unpacked` holds the files Studio actually loads, extracted from it. This is the one people miss, and it is the one that blocks a reinstall: Studio will not re-extract over an `Unpacked` folder that is already there. Note **Two things that catch people out.** `AppData` is hidden by default, so you cannot browse to it – paste the whole path into the File Explorer address bar instead and press Enter. You may also see these written as `%AppData%` and `%LocalAppData%`. Those are just Windows shortcuts for `C:\Users\\AppData\Roaming` and `C:\Users\\AppData\Local`, and you can paste them into the address bar in place of the first part of the path if you prefer – they save typing your user name. Note **Quick way to check:** paste each path into the Windows Run dialogue (`Win+R`) or File Explorer address bar. If the folder exists and contains an old `Supervertaler for Trados.sdlplugin`, delete it. Also check for an `Unpacked\Supervertaler for Trados` folder at the same level and delete it if present. After removing the old files, double-click the new `.sdlplugin` to install it fresh, then start Trados. #### Uninstalling To remove the plugin: 1. Open Trados Studio 2. Go to **Help > Plugin Management** 3. Find “Supervertaler for Trados” in the list 4. Click **Uninstall** 5. Restart Trados Studio *** #### Next Steps * [Getting Started](/trados/getting-started/) – set up your first termbase and API key # Keyboard Shortcuts All keyboard shortcuts available in Supervertaler for Trados, with Mac equivalents for users running Trados in Parallels. Note **Mac users (Parallels):** Ctrl = Control, Alt = Option on Mac keyboards. Check **Parallels → Preferences → Shortcuts** if your modifier key mapping differs. ## First-time setup: free up Trados shortcuts Trados Studio ships with its own default bindings on several of the key combinations Supervertaler uses – and the Trados binding wins. If a Supervertaler shortcut does nothing, this is almost always why. Go to **File → Options → Keyboard Shortcuts**, search for the Trados action named below, and delete (or reassign) its binding: | Shortcut | Trados default action (delete its binding) | Supervertaler action that needs it | | ------------ | ------------------------------------------ | ------------------------------------ | | `Ctrl+Alt+T` | Insert TM Symbol (™) | Add term entry (full editor) | | `Ctrl+Alt+N` | New Cloud Project | Quick-add non-translatable term | | `Ctrl+Alt+G` | Open GroupShare Project | Auto-tag active segment (AutoTagger) | | `Alt+Up` | Focus Previous Row | Quick-add term to project termbase | | `Alt+Q` | Tell me what you want to do | Open QuickLauncher | Note The same applies to any other Supervertaler shortcut that appears dead: search **File → Options → Keyboard Shortcuts** for that key combination and clear the Trados binding. You only need to do this once per Trados installation – but repeat it after reinstalling or resetting Trados Studio. Note **Also running Supervertaler Workbench?** The two products share several shortcuts deliberately – `Alt+Down`, `Ctrl+Alt+T`, `Alt+1`…`Alt+9` do the same job in each – and that is safe, because whichever window you are working in responds. The exception is Workbench’s handful of **global** hotkeys, which fire whatever application is in front, Trados included. None of them clash with the shortcuts below; see [Workbench global hotkeys](/workbench/settings/shortcuts/) if you rebind one. ## Terminology | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | ----------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `Alt+Down` | `Option+Down` | Quick-add term to write termbases | | `Alt+Up` | `Option+Up` | Quick-add term to project termbase | | `Ctrl+Alt+T` | `Control+Option+T` | Add term entry (opens full editor with definition, domain, notes, URL, client, project, and synonyms) | | `Ctrl+Alt+A` | `Control+Option+A` | [Add term with abbreviation](/trados/termlens/adding-terms/) – the AI fills in the term pair and both abbreviations from the segment, for you to confirm (from v18/19.20.157) | | `Ctrl+Alt+N` | `Control+Option+N` | Quick-add non-translatable term | | `Ctrl` (tap) | `Control` (tap) | Toggle the floating **TermLens popup** (open / close) | | `Alt+P` | `Option+P` | Open **TermPicker** (list-based) – or focus the TermPicker pane when it is docked | | `Alt+1` … `Alt+9` | `Option+1` … `Option+9` | Insert term 1–9 by badge number | ## AI Translation | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | -------------- | --------------------------------------------------------------------------------------------------------------------- | | `Alt+Q` | `Option+Q` | Open QuickLauncher prompt menu | | `Alt+T` | `Option+T` | Translate the active segment (uses Batch Translate settings, incl. document + SuperMemory context from v18/19.20.149) | Note **QuickLauncher moved from `Ctrl+Q` to `Alt+Q` in v18/19.20.184.** `Ctrl+Q` is a Trados factory default (“View Internally Source”) and Trados wins, so QuickLauncher did nothing on a fresh install until you went into Options and cleared that binding – with no error to tell you why. `Alt+Q` fits the `Alt`+letter family the other shortcuts use – but Trados also puts **Tell me what you want to do** on it (confirmed in Studio 2024), so it is in the table above: clear that binding once and QuickLauncher works. Tell Me is the search box at the top right of the ribbon and stays clickable without its shortcut. If you had already cleared the Trados binding to make `Ctrl+Q` work, that key will now do nothing for QuickLauncher – use `Alt+Q`, or reassign it in **File → Options → Keyboard Shortcuts**. **Why `Alt+T` and not `Ctrl+T`?** `Ctrl+T` is a Trados factory default (“Apply Translation Result”). Binding *both* to one key made a single press fire both commands, which raced on the same segment and could freeze Studio – so the default moved to the collision-free `Alt+T` (in plugin v18/19.20.119). If you upgraded from an earlier version and still have it on `Ctrl+T`, reassign it to `Alt+T` (or any free key) in **File → Options → Keyboard Shortcuts**; Studio keeps your existing binding across updates. ## Voice Commands | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | ------------------ | --------------------------------------------------------------------------------------------------------------- | | `Ctrl+Alt+D` | `Control+Option+D` | Toggle [voice commands](/trados/voice-commands/) on/off (same as clicking the 🎤 button in the TermLens header) | Note **Why `Ctrl+Alt+D` and not `Ctrl+Alt+V`?** It was `Ctrl+Alt+V` until v18/19.20.157. Supervertaler Workbench uses that combination for its own voice-command push-to-talk, and registers it as a **global** hotkey – one that fires whichever application is in front, Trados included. If you run both, a single press started two listeners; and because the Workbench one is a *hold* while this one is a *toggle*, releasing the key stopped only Workbench’s and left this one running with nothing visible having switched it on. If you upgraded and still have it on `Ctrl+Alt+V`, reassign it in **File → Options → Keyboard Shortcuts** – Studio keeps your existing binding across updates. ## QuickLauncher Shortcuts | Shortcut (Windows) | Shortcut (Mac) | Action | | --------------------------- | --------------------------------------- | --------------------------------------------- | | `Ctrl+Alt+1` … `Ctrl+Alt+9` | `Control+Option+1` … `Control+Option+9` | Run QuickLauncher prompt assigned to slot 1–9 | | `Ctrl+Alt+0` | `Control+Option+0` | Run QuickLauncher prompt assigned to slot 10 | Note Assign prompts to slots in **Settings → Prompts**. Select a QuickLauncher prompt and choose a shortcut from the dropdown in the detail pane. If no slots are assigned, the shortcuts default to menu position order. ## SuperMemory | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | ------------------ | -------------------------------------------------------------- | | `Ctrl+Alt+M` | `Control+Option+M` | Quick Add – add a term or correction to the active memory bank | ## SuperSearch | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | -------------- | ---------------------------------------------------------------------------------------- | | `Alt+S` | `Option+S` | Open SuperSearch – searches for the selected source/target text across all project files | | `Alt+W` | `Option+W` | Search the web for the selected term, in your project’s language pair | ## AutoTagger | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | ------------------ | ----------------------------------------------------------------------------------- | | `Ctrl+Alt+G` | `Control+Option+G` | Auto-tag the active segment – places the source segment’s inline tags in the target | ## Navigation and Display | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | -------------- | ----------------------------------------------------------------------------------------------------------------- | | `F1` | `F1` | Context-sensitive help | | `F2` | `F2` | Expand selection to word boundaries | | `F5` | `F5` | Force-reload Supervertaler termbases from disk and refresh TermLens display (does not reload MultiTerm termbases) | ## Shortcuts for Terms 10+ When a segment has more than 9 matched terms, you can still insert terms by number using Alt+digit. TermLens offers two shortcut styles – choose the one you prefer in **Settings > TermLens > Term shortcuts**. ### Sequential (default) Type the term number digit by digit. Each badge shows the plain term number (10, 11, 12, …). | Shortcut (Windows) | Shortcut (Mac) | Inserts | | ------------------ | -------------- | ------- | | `Alt+10` | `Option+10` | Term 10 | | `Alt+23` | `Option+23` | Term 23 | | `Alt+45` | `Option+45` | Term 45 | After the first digit, TermLens waits briefly for a possible second (or third) digit. If no further digit is pressed, the single-digit term is inserted. ### Repeated digit Press the **same digit key** multiple times. Each badge shows the repeated digit (11, 222, 3333, …). | Presses | Windows | Mac | Badge | Terms | | ------- | ------------------------- | ------------------------------- | --------------------- | ----- | | 1x | `Alt+1` … `Alt+9` | `Option+1` … `Option+9` | **1** – **9** | 1–9 | | 2x | `Alt+11` … `Alt+99` | `Option+11` … `Option+99` | **11** – **99** | 10–18 | | 3x | `Alt+111` … `Alt+999` | `Option+111` … `Option+999` | **111** – **999** | 19–27 | | 4x | `Alt+1111` … `Alt+9999` | `Option+1111` … `Option+9999` | **1111** – **9999** | 28–36 | | 5x | `Alt+11111` … `Alt+99999` | `Option+11111` … `Option+99999` | **11111** – **99999** | 37–45 | Note In both modes, when a segment has 9 or fewer matches, pressing Alt+N inserts immediately with no delay. Note Terms beyond 45 have no keyboard shortcut. Use the **TermLens popup** (tap `Ctrl`) or **TermPicker** (`Alt+P`) to insert them. *** ## See Also * [TermLens](/trados/termlens/) * [Supervertaler](/trados/ai-assistant/) * [SuperSearch](/trados/supersearch/) * [AutoTagger](/trados/autotagger/) * [Batch Translate](/trados/batch-translate/) * [SuperMemory](/trados/ai-assistant/super-memory/) # Licensing Supervertaler for Trados uses a simple subscription model: one product, one price, everything included. ## Free Trial When you first install Supervertaler for Trados, a **14-day free trial** starts automatically. During the trial, all features are unlocked – TermLens, AI Assistant, SuperSearch, memory banks, Studio Tools, and everything else. No sign-up or credit card is required to start the trial. The remaining days are shown in the **Licence** tab in Settings and in the About dialogue. ## Pricing | | Monthly | Annual | | ---------------------------- | --------- | --------- | | **Supervertaler for Trados** | €20/month | €200/year | One plan, all features included: TermLens inline terminology, AI Assistant & Batch Translate, SuperSearch cross-file search & replace, SuperMemory memory banks, Studio Tools, Clipboard Mode, QuickLauncher, Prompt Library, MultiTerm support, Incognito Mode, and all future features. Note Annual plans include **2 months free** compared to monthly billing. ## Purchasing a Licence 1. Visit [supervertaler.com/trados](https://supervertaler.com/trados/) and click **Subscribe** 2. Complete the checkout – you will receive a **licence key** by email 3. Open Trados Studio → **Settings → Licence** tab 4. Paste your licence key and click **Activate** Your licence allows activation on up to **2 machines** (e.g. a desktop and a laptop). ## Activating Your Licence 1. Open Trados Studio 2. Click the **gear icon** (⚙) on the TermLens or Supervertaler Assistant panel 3. Go to the **Licence** tab 4. Enter your licence key in the text field 5. Click **Activate** A confirmation message appears when activation succeeds. The Licence tab shows your plan name, masked licence key, status, and last verification date. Tip You can also reach the Licence tab by clicking the licence status text in the **About** dialogue (accessible via the **?** button on any panel). ## Managing Your Subscription From the **Licence** tab in Settings, you can: * **Verify Now** – manually check your licence status with the server * **Deactivate** – remove the licence from this machine (frees up an activation slot) * **Manage subscription →** – opens the Lemon Squeezy billing portal where you can update payment details or cancel ## Offline Use After activation, the plugin caches your licence status locally. You can work offline for up to **30 days** before the plugin needs to verify your licence again. When you reconnect to the internet, verification happens automatically in the background. ## What Happens When the Trial Expires After the 14-day trial ends: * **No licence** – all features show a “licence required” overlay. Your termbases, settings, and prompt library are preserved. * **Active licence** – all features are unlocked. Activating a licence immediately unlocks all features. ## Changing Machines If you replace a computer or need to move your licence: 1. On the old machine: open **Settings → Licence** and click **Deactivate** 2. On the new machine: enter your licence key and click **Activate** If you can no longer access the old machine, the activation slot will be freed automatically when the licence is next validated. ## Privacy & Security The plugin makes **no network calls** except to: 1. **Your chosen AI provider** (OpenAI, Anthropic, Google Gemini, OpenRouter, or local Ollama) – only when you use AI features 2. **Lemon Squeezy licence API** (`api.lemonsqueezy.com`) – for licence activation and periodic validation 3. **Anonymous usage statistics** (strictly opt-in) – if you consent, a single ping on startup sends only: plugin version, OS version, Trados version, and system locale. See [Usage Statistics](/trados/settings/usage-statistics/) for details. The licence validation sends only your licence key and a hashed machine fingerprint (a one-way hash of your computer name and Windows user ID). No personal data, no translation content, no termbase information is ever collected. Your API keys are stored locally in `%LocalAppData%\Supervertaler.Trados\settings.json` and are never transmitted anywhere except to your chosen AI provider. Note The full source code is available on [GitHub](https://github.com/Supervertaler/Supervertaler-for-Trados) for security audit. You can verify exactly what the plugin does and does not transmit. # Supervertaler MCP Server The Supervertaler MCP Server connects **Claude Desktop** directly to your live Trados Studio session. You chat in Claude’s own window, and it answers from your real project data: the document open in the editor, your translation memories, and your termbases. It can also make changes for you, always under your supervision. > **Which AI apps work?** Any app that can run a **local (STDIO) MCP server on your own machine**. Claude Desktop is the easiest, because the plugin ships a one-click extension for it. **ChatGPT’s desktop app works too** *(confirmed August 2026)*, as do **Google Antigravity** *(confirmed September 2026)*, Claude Code, Gemini CLI and **Mistral Vibe CLI** *(confirmed August 2026)* – see [Setting it up](#setting-it-up) for each. Antigravity is the Google-side equivalent of Claude Desktop, and the way to drive Trados from Gemini. What cannot work is anything that runs the server in the cloud rather than on your PC. That rules out the claude.ai and chatgpt.com **websites**, and also Mistral’s **Vibe web app**, whose custom MCP connectors accept only a remote `https://` URL. The Supervertaler bridge is local by design, so your project never leaves your machine, and a cloud-hosted client has no route to it. > > *Earlier versions of this page said ChatGPT desktop could not be used. That was true when written and is no longer: the desktop app has since added support for local STDIO servers.* MCP ([Model Context Protocol](https://modelcontextprotocol.io/)) is the open standard that lets AI applications securely call tools exposed by other programs. The Supervertaler MCP Server is the first MCP server that talks to a **live** Trados Studio editor session – other Trados-related MCP servers work on project files on disk, not the document you are working on. **Prefer to watch?** The [MCP Server screencast](https://www.youtube.com/watch?v=hKQ62_IU0fk) (9 min) shows it in use: Trados Studio 2026 driven from Claude’s chat window. [Supervertaler MCP Server – controlling Trados Studio 2026 from Claude Chat](https://www.youtube.com/embed/hKQ62_IU0fk)  **Claude Code.** Asking for a glossary drawn from the live Trados Studio project – it reads the open document and checks the user's termbase, then answers in chat.  **And from a phone.** The same project, asked from Claude's Android app: the conversation is attached to Claude Desktop on the PC, where the MCP server runs, so the tools reach Trados Studio just as they do on the desktop. The PC has to stay on with Studio open – but you do not have to be at it. ## What you can ask With Trados Studio open and a document in the editor, you can ask your AI assistant things like: * “What’s the status of my Trados project? How many segments are left?” * “How many times does the word *doekrol* appear in my project, and did I translate it consistently?” * “Find all Draft segments containing *flange* and show me the translations.” * “How did I translate this phrase before?” (searches your Supervertaler TMs) * “What does my termbase say for *sluitkracht*?” * “Draft translations for the untranslated segments and set them to Draft so I can review them.” * “We agreed *draagarm* = *support arm* – add it to my termbase.” * “Work with the 2026 one.” (when you have two versions of Trados Studio open – see [Two Studios open at once](#two-studios-open-at-once-from-v1820184)) Unlike the [AI-friendly bilingual export](/trados/import-export/) workflow, there is no export/re-import cycle: the AI reads the live document on demand, and changes it makes appear in Studio while you chat. ## What the AI can do The server exposes these tools to the AI app: | Tool | What it does | | ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `help` | A curated menu of what you can ask – shown when you say *“what can I do?”* *(v18.20.106)* | | `session_report` | How many bytes of tool results this session has sent the AI, per tool, biggest first – so you can see which tools are filling (and re-billing) the conversation *(v18.20.158)* | | `list_trados_instances` | Which Trados Studios are running, the project open in each, and which one the AI is talking to – for when you have more than one open *(v18.20.184)* | | `select_trados_instance` | Tell the AI which of several running Studios to work with, by Studio version or project name *(v18.20.184)* | | `get_active_project` | Project name, language pair, active file, segment counts per confirmation status | | `get_segments` | List segments, with filters (status, contains-text, file) and paging – or fetch exact segments by the grid number(s) you see in Studio (`fromNumber`/`toNumber`) *(grid numbers v18.20.114)* | | `get_files` | The files of a merged multi-file document, with per-file segment counts *(v18.20.95)* | | `get_active_segment` | The segment you are editing right now, with TM matches and termbase hits | | `get_project_statistics` | Analysis bands and per-file confirmation statistics – word counts, progress *(v18.20.95)* | | `search_studio_tm` | Concordance-search the Trados TMs attached to the project (.sdltm and GroupShare) *(v18.20.95)* | | `lookup_term` | Look up a term in your termbases (exact first – against source and target terms alike – then substring matching); hits report which column matched and are never reoriented *(v18.20.153)* | | `find_inconsistencies` | Repeated source segments whose translations differ *(v18.20.95)* | | `compare_document_to_tm` | Every segment where your translation differs from what the TM already holds for the same source – the pre-delivery consistency check *(v18.20.148)* | | `check_numbers` | Translated segments whose numbers differ between source and target *(v18.20.95)* | | `check_tags` | Translated segments with missing or extra inline tags – compares underlying tag ids as well as counts, so two tags sharing one id are caught *(ids from v18.20.157)* | | `check_terminology` | Translated segments that don’t use the termbase’s expected translation – longest-match-wins, ranked by signal rather than raw count, restrictable to a curated termbase *(overhauled v18.20.157)* | | `check_nbsp` | Translated segments that lost a non-breaking space the source had – invisible on screen, so nothing else catches it *(v18.20.148)* | | `get_coverage` | Which segments have been neither written nor explicitly reviewed this session, per TM match band – so “the fuzzy band was read” becomes checkable instead of remembered *(v18.20.157)* | | `mark_reviewed` | Record that segments were read source-against-target and deliberately left unchanged – session-scoped, never written to the file *(v18.20.157)* | | `get_tracked_changes` | The document’s tracked changes as (before, after) pairs per segment – how you corrected the drafts – optionally saved into the active SuperMemory bank’s reference folder for future projects *(v18.20.158)* | | `list_resources` | The TMs and termbases attached to your project and Supervertaler setup – per termbase with its Read/Write ticks and whether it is the Project termbase *(roles from v18.20.159)* | | `list_projects` | Every project registered in Trados Studio – across Studio 2026/2024/2022 – with status and paths *(v18.20.111)* | | `get_project` | Details of any registered project by name, without opening it *(v18.20.111)* | | `list_tms` | The file TMs on this machine (Studio folders + project references) *(v18.20.111)* | | `list_project_templates` | Your Trados project templates *(v18.20.111)* | | `update_segments` | Write translations and/or set confirmation statuses (see safety rails below) | | `add_term` | Add a term pair to your Write termbases – direction-aware per termbase, with optional definition/domain/notes, termbase targeting by name or by role (project vs background – see below), and a per-termbase echo of exactly what was stored; duplicates echo the existing entry they matched *(scope + duplicate echo v18.20.159)* | | `update_term` | Fix an existing entry in your Write termbases – exact-match; can change the term pair and, from v18.20.159, also the entry’s notes, definition and domain – only the fields you name change, everything else is preserved | | `delete_term` | Remove an entry from your Write termbases – destructive, so the AI confirms first *(v18.20.113)* | | `import_project_termbase` | Copy the Trados termbase attached to the open project (`.sdltb` / `.ttb`) into a Supervertaler termbase in one step – the same job as the *Import .sdltb/.ttb…* button. Ask for a dry run first and it reports the entry count, language pair and field mapping before writing anything; running it twice adds nothing the second time. An existing destination must be Write-enabled, a new name is created for you, and your Trados termbase is only ever read *(v18.20.175)* | | `insert_into_active_segment` | Insert text into the active segment’s target (like Apply-to-target) | | `save_document` | Save the open document (Ctrl+S) – only when you ask or approve *(v18.20.115)* | | `go_to_segment` | Move the Studio editor to a specific segment (by grid number or id) | | `find_and_replace` | Find & replace across the target text – tag-safe, with a preview before applying | | `get_comments` | Read the Trados comments in the document | | `add_comment` | Add a Trados comment to a segment (flag a source issue, leave a review note) | | `update_comment` | Edit an existing Trados comment – its text, its severity, or both *(severity from v18.20.159)* | | `delete_comment` | Remove a Trados comment (or all of a segment’s) – destructive, so the AI confirms first *(v18.20.116)* | | `run_verification` | Run Studio’s Verify Files (QA Checker) and return the findings per segment – flagged as stale if the AI has unsaved edits *(stale flag v18.20.148)* | | `analyze_files` | Run **Analyse Files** – computes the perfect/exact/fuzzy/new/repetition leverage breakdown *(v18.20.106)* | | `pretranslate` | Run **Pre-translate Files** – fill untranslated segments with their TM matches | | `update_tm` | Run **Update Main Translation Memories** – write confirmed segments to the project TM | | `export_target` | Run **Generate Target Translations** – write out the translated target files | | `get_task_status` | Check a background batch task’s progress (analyse, pre-translate, …) *(v18.20.104)* | | `list_prompts` | Browse your Supervertaler prompt library, optionally filtered by folder or search term *(v18.20.101)* | | `get_prompt` | Read the full text of one of your prompts *(v18.20.101)* | | `save_prompt` | Create a new prompt, or update one of your own – built-in defaults are protected *(v18.20.101)* | | `get_prompt_context` | Everything the AI needs to write a translation prompt tailored to your open project – source text, domain, terms, TM examples *(v18.20.109)* | | `get_supermemory_context` | The active [SuperMemory](/trados/ai-assistant/super-memory/) bank for this project – its brief, terminology table and style rules, plus the `_shared` bank of house defaults that the client bank overrides *(three-file banks from v18.20.169)* | | `search_supermemory` | Search your active memory bank **and `_shared`** by keyword – *“what did I decide about this term, and why?”*. Each hit says which bank it came from *(v18.20.146; `_shared` included from v18.20.172)* | | `list_supermemory_banks` | Your memory banks and which one is active. `_shared` is labelled as the always-loaded layer rather than listed as an ordinary bank *(v18.20.146; roles from v18.20.172)* | > The four batch tasks (`analyze_files`, `pretranslate`, `update_tm`, `export_target`) run in the **background** and return immediately – the AI polls `get_task_status` and tells you when they finish, so a long analysis never stalls the chat. > > This list grows over time. For the current set on your installed version, just ask the AI *“what can I do?”* (the `help` tool). ### Safety rails on write actions * Translations written by the AI are set to **Draft** status unless it explicitly sets another status – so you can filter for them in Studio and review everything. * **Locked segments are never touched.** * **Find & replace keeps each segment’s confirmation status** *(from v18.20.148)*. Editing a segment’s content normally demotes it to Draft, which meant a single consistency sweep over a finished file could quietly leave thousands of segments unconfirmed. The AI can still ask for a specific status when you want one. * Updates are limited to 40 segments per call; larger jobs are processed in reported batches. *(Lowered from 200 in v18.20.148: bigger batches could outlast the connection timeout, and because the write had already gone through, the AI couldn’t tell success from failure.)* * Changes land in the open document but are **not saved automatically** – saving stays your decision. From v18.20.115 the AI can run the save for you (`save_document`, same as Ctrl+S), but only when you ask or approve – *“save and run the analysis”* is one instruction, silent saving is not allowed. * The AI is instructed to only make changes you asked for, and to report exactly what it changed. ### Two Studios open at once *(from v18.20.184)* Trados Studio 2024 and Trados Studio 2026 can run side by side, each with its own project and its own connection. Two AI apps can then work on **two different projects at the same time** – ChatGPT drafting one job while Claude Desktop drafts the other, in two Studio windows, on one machine. Each Studio announces itself with its version and the project it has open, so an AI app can tell them apart and say which one it is working in. **Reading is always answered**, and the reply says which Studio and project it came from – so an answer about the wrong project is obvious rather than silent. **Changing anything is refused until you say which Studio you mean.** Anything that edits segments, terms, comments or files stops and lists the Studios that are open, instead of picking one. This is the point of the feature: writing into the wrong project is the one mistake you cannot see happening. **Say which one you want** and editing is enabled again for the rest of the chat: * *“Work with the 2026 one.”* * *“Use the Acme project.”* * *“Which Trados instances are running, and which are you using?”* – if you want to see the list first. The choice follows **the project, not the process**, so it survives that Studio being closed and reopened. You do not have to say it again after a restart. Closing the other Studio works just as well: the remaining one is unambiguous immediately, with nothing to restart. #### Translating two projects at once 1. Open both Studios, each with its project. 2. In the first AI app: *“Use the 2024 one”*, then set it going. 3. In the second AI app: *“Use the 2026 one”*, then set it going. Each app now edits only its own project and cannot touch the other’s document. Both can work at the same time. One assistant still works in **one** project at a time – this is two conversations running in parallel, not one assistant spanning both. #### Pinning an app to one Studio permanently If you always pair the same app with the same Studio, set it once in that app’s own MCP configuration instead of saying it each time. Add `--instance` to the server’s `args`: ```json "args": ["--instance", "2024"] ``` or set the environment variable `SUPERVERTALER_TRADOS_INSTANCE` to `2024`, `2026`, or part of a project name. A pinned app never asks – and if the Studio it wants is not running, it says so rather than quietly using the other one. #### Two chats in the same app – read this before switching windows The recipe above uses **two different AI apps** for a reason. One app runs **one** Supervertaler server, and *“Use the 2026 one”* is remembered **by that server** – not by the chat you said it in. So if you open two Claude Desktop chats, tell one *“use 2026”* and the other *“use 2024”*, and then switch between them, **both are now pointed at whichever Studio you named last.** Switching windows does not switch Studios, and the refusal above will not catch it: it only fires when *nothing* has been chosen, not when the choice is stale. This is exactly how a translation could land in the wrong project. **Give each chat one line that names its Studio, and it works only in that one.** In Claude Desktop, *Projects → Project instructions*, put in the 2026 chat’s project: > This project works only with Trados Studio 2026. Pass `instance: "2026"` on **every** Supervertaler tool call – reads as well as writes. Do **not** call `select_trados_instance`; the `instance` argument replaces it. and `"2024"` in the other. From v18.20.190 every Supervertaler tool takes an optional `instance` – `"2024"`, `"2026"`, or part of the project name – resolved against the Studios actually running. A call scoped to 2026 is answered by, or written to, the 2026 Studio and no other; one that names a Studio which is not running is **refused by name**, nothing touched. Because each call carries its own answer, the two chats never share a setting and never need `select_trados_instance` – which is the trap to avoid, since that command sets one selection shared by *every* chat, so two chats using it just overwrite each other. Pass `instance` on every call and that whole problem disappears. This is the reliable way, and it needs no second server and no file to edit. Every write also now **says which Studio and project it landed in**, in its reply, so even without the `instance` line a translation that went to the wrong place is visible at once rather than discovered later. **If you would rather it be impossible than declared** – a second server, pinned, so the two can never even see each other. Keep the extension for one Studio and add a second server for the other in `%APPDATA%\Claude\claude_desktop_config.json`: ```json { "mcpServers": { "supervertaler-2024": { "command": "C:\\Users\\\\Supervertaler\\mcp\\SupervertalerMcpServer.exe", "args": ["--instance", "2024"] } } } ``` Each chat’s instructions then name the server to use. Every tool appears twice, prefixed by server name; here that is the point, not a mistake. For most people the `instance` line above is simpler and just as safe. > The Connect dialog warns when a second Studio is running and names its project, since there is no way to tell from inside the first one. ### Direction-aware termbase writes *(from v18.20.153)* Termbases have a declared language direction, and yours don’t all point the same way – a main termbase might be en→nl while a project termbase is nl→en. From v18.20.153 `add_term` handles this per termbase: * **The AI states which language each side of the pair is in** (`sourceLang`/`targetLang`), and every termbase written to stores the pair according to **its own** declared direction. One request fills an en→nl and an nl→en termbase correctly at the same time – each entry the mirror of the other. * **Ambiguity refuses instead of guessing.** If the languages can’t be established – no document open, or a termbase whose language pair doesn’t match – the write is refused with an explanation rather than performed silently. There is deliberately no language detection: technical term pairs are routinely identical in both languages (*radar*, *transponder*), so a detector would guess, and a wrong entry that *looks* fine is worse than a refusal. * **The response proves what happened.** Every targeted termbase reports back individually: added (echoing exactly what was stored, in stored order, with a flag when the pair was reoriented), already present, or refused with the reason. `lookup_term` returns entries exactly as stored – never reoriented – and says which column your query matched, so a write can always be independently verified. * **Entries can carry their context.** `add_term` accepts a definition, domain and notes alongside the pair, and can be told to write to specific termbases only instead of all Write-enabled ones. ### Project vs background termbases *(from v18.20.159)* A common setup is two Write-enabled termbases with different roles: a large personal **background** termbase that accumulates terminology across all clients, and a per-job **project** termbase (the one with the **Project** tick in the Supervertaler Termbases settings – the same tick that renders its hits pink in TermLens). Before v18.20.159 a plain “add this term” wrote to both, indiscriminately. Now the roles are first-class: * **The AI can target a role instead of a name.** *“Add this to my project termbase”* writes only to the Project-ticked termbase; *“add this to my background termbase”* writes only to the Write-enabled ones without the tick. Job-specific decisions no longer silently pollute your general glossary. Asking for the project termbase when none is ticked is refused with an explanation, never silently redirected. * **Every result names its role.** Each termbase in an `add_term` response reports whether it is `project` or `background`, and `lookup_term` hits carry the flag too – so the AI can weight your curated project decisions above general-glossary entries when they conflict. * **Duplicates show what they matched.** When a termbase refuses a pair as already present, the response now includes the existing entry (its id and exact stored pair) instead of a bare “duplicate” – no more guessing what it collided with. * **Stale project termbases are flagged.** If the Project-ticked termbase’s name shares no word with the open project’s name – the classic sign of a tick left over from a previous job – the `add_term` response says so, before weeks of terms land in the wrong client’s termbase. ## Prompt cookbook You talk to the AI in plain language – there are no commands to memorise. The AI decides which tools to call from what you say. This section lists, per task, the kinds of things you can say, so you know the full range of what’s possible. Mix and combine freely (“find X, then fix Y”). > **Not sure where to start? Just ask *“What can I do?”*** (or *“what can you do?”*) and the assistant shows a grouped menu of everything below – so you don’t have to read this page first.  **Claude Desktop.** Ask *"What can I do?"* and the assistant lists what you can ask it, grouped by task. ### Project status and progress * “What’s the status of my Trados project?” * “How many segments are left to translate?” * “Which file am I working on, and what’s the language pair?” * “How many words are still untranslated?” / “Give me the analysis statistics – fuzzies, repetitions, new words.” *(from v18.20.95)* * “How far along is each file in this project?” *(from v18.20.95)* * “What projects do I have?” / “When did I create the ACME job, and where is it on disk?” *(all Studio versions’ registries – from v18.20.111)* * “Which TMs and project templates are on this machine?” *(from v18.20.111)* ### Finding and reading segments * “Show me all untranslated segments.” * “Show me the Draft segments so I can see what the AI wrote earlier.” * “Find all segments containing *flange*.” * “How many times does *doekrol* appear in this project? Is it translated consistently?” * “Show me segments 50 to 100.” (paging) * “List the files in this merged document.” / “Only show me segments from the contract file.” *(from v18.20.95)* ### Terminology * “What does my termbase say for *sluitkracht*?” * “Look up *support arm* – do I have an established translation?” * “Go through the project and make me a glossary of the key terms.” * “We agreed *draagarm* = *support arm* – add it to my termbase.” * “Extract the recurring technical terms from this document and add the ones I approve to my termbase.” * “That pair is outdated – replace it with the official MDR term in both termbases.” *(update, exact-match, audited in chat – from v18.20.113)* * “Delete that junk entry the QA keeps flagging.” *(Write-enabled termbases only; the AI confirms before deleting – from v18.20.113)* * “Only consult my **active** termbases for this lookup.” *(restricts to termbases with Read ticked; otherwise inactive hits are flagged – from v18.20.113)* * “Add *commandovoering* = *command and control* to my termbases – with the NATO definition and a usage note.” *(direction-aware per termbase, with definition/domain/notes – from v18.20.153)* * “Add this pair to the Acme termbase only.” *(write to named termbases instead of all Write-enabled ones – from v18.20.153)* * “Add this to my **project** termbase only.” / “Put that one in my **background** termbase, it’s not client-specific.” *(role-based targeting via the Project tick – from v18.20.159)* * “Extend the usage note on *eenzelfde* with a warning about the split spelling.” *(edit an entry’s notes, definition or domain in place, without touching the pair – from v18.20.159)* ### Translation memory * “How did I translate this sentence before?” *(searches the Trados TMs attached to your project – from v18.20.95)* * “Search my TM for *scherminrichting*.” * “Search only the target side of my TMs for *roller blind*.” *(from v18.20.95)* * “Before translating, check my TM and termbase and follow what you find.” ### The segment I’m working on * “Translate this segment.” / “Explain this sentence.” * “What do my TM and termbase say about the current segment?” * “Give me three alternative translations for this segment, then insert the one I pick.” ### Writing translations (always reviewable) * “Draft translations for all untranslated segments – I’ll review them in Studio.” * “Translate the segments containing *warranty*, use my termbase, set them to Draft.” * “Redo segment 14 – too literal, make it flow better, then update it.” * “Set all my Draft segments to Translated.” (status-only changes work too) Everything the AI writes lands as **Draft** unless you say otherwise, locked segments are never touched, and nothing is saved until you save in Studio. ### Quality and consistency * “Find segments where the source and target numbers don’t match.” *(from v18.20.95)* * “Check my tags – any segments missing formatting?” *(from v18.20.95)* * “Check my translated segments against the termbase and list violations.” *(from v18.20.95)* * “Find all repeated sentences that I translated differently.” *(from v18.20.95)* * “Check whether I’ve lost any non-breaking spaces.” *(from v18.20.148)* * “Put a non-breaking space between every value and its unit.” *(from v18.20.148)* * “Where does my translation differ from the client’s reference TM?” *(from v18.20.148)* * “Run all your QA checks and give me a report.” * “…then align them all to the best version.” (pairs with the write tools) Comparing against the TM is worth a note of its own, because it answers a different question from concordance search. Searching the TM tells you whether a phrase *you already suspect* was translated before – one query at a time, for things you thought to look up. `compare_document_to_tm` goes the other way: it reads the attached TMs once and checks **every** segment whose source appears in them, then reports only where your wording differs. That surfaces the cases you had no reason to check, which is exactly where an established client rendering gets missed. Two things to keep in mind. A difference is not automatically a mistake – on a real job most of them are deliberate improvements, and an improvement is indistinguishable from an error here, so the assistant is instructed to present the list for you to judge rather than align anything itself. And only segments whose source matches the TM word for word are compared, so a clean result means “nothing contradicts the TM”, not “the whole document agrees with it”. Non-breaking spaces deserve a note of their own. They are invisible everywhere – in Studio, in the AI’s view of your segments, in any report – so a lost one usually surfaces only when the client rejects the file. That matters if your style guide asks for one between a value and its unit (230 V, 3,5 mm, 50 %) or before a figure reference. `check_nbsp` compares each translated segment against its source and lists the ones that came out with fewer. Inserting them used to be the harder half. A non-breaking space typed straight into a tool call survives the trip only sometimes: depending on the AI client and the individual call, it either arrives intact or is normalised to an ordinary space on the way, and because the write itself succeeds, nothing tells you which happened. Escape codes don’t help either – the AI client decodes them into the character first, and then the character is what gets flattened. Being intermittent makes it worse rather than better: it works when you try it, and fails on the job. From v18.20.148 the AI can write the character as the HTML entity ` `, which Supervertaler turns into a real non-breaking space at the Trados end. Plain ASCII travels intact, so nothing en route can mangle it. This works for both writing translations and find & replace, and searching, so *“put a non-breaking space between every value and its unit”* fixes a whole document in one pass – and because find & replace preserves confirmation status, doing that to a finished file leaves it finished. It is deliberately opt-in, so a document that genuinely contains the text ` ` (an HTML manual, say) is never silently rewritten. ### Resources * “Which TMs and termbases is this project using?” *(from v18.20.95)* * “Are any of my termbases actually switched on for this project?” *(from v18.20.148)* – termbases are enabled per project, so a project with all of them off looks exactly like one with no terminology at all. The AI is now warned when nothing is read-enabled instead of silently finding no terms. ### Your memory bank *(from v18.20.146)* Your [SuperMemory](/trados/ai-assistant/super-memory/) memory bank holds the reasoning behind your decisions – why a term was chosen, what a client insists on, what you rejected last time – which is exactly what a TM or termbase cannot tell an AI. From this version the AI can read it over MCP, not just inside the Supervertaler chat panel: * “Read my memory bank for this project before we start.” * “What did I decide about *inrichting* for this client, and why?” * “Check this translation against my style guide and terminology notes.” * “Which memory bank are you using?” The AI cites the articles it drew from by path, so you can open them in Obsidian and check its reasoning. Retrieval is read-only – nothing is written back to the bank over MCP. If you have turned memory-bank context off under Settings → AI Settings, these tools stay quiet too. A large bank will not fit into one answer, so some of it is left out – and **the AI is now told which files those were** *(from v18.20.183)*. It used to be trimmed silently, which is the worse failure: two of your three articles look exactly like all three, so a rule you had written down could be absent from the answer with nothing to say so. If the AI mentions that something was left out, ask it to read the bank again with a larger budget, or ask about that file by name. ### Your prompt library *(from v18.20.101)* The AI can read and improve the Markdown prompts in your Supervertaler prompt library – the same ones you use in the QuickLauncher and Batch Translate (and shared with Supervertaler for memoQ): * “List my prompts.” / “Show me the prompts in my Translate folder.” * “Show me my Default Translation Prompt.” * “Look at my Default Translation Prompt and suggest improvements for patent work, then save it as a new prompt.” * “Turn what we just worked out into a prompt and save it as *Client X house style*.” * “Look at my open project and write me a translation prompt tailored to it.” *(from v18.20.109 – the AI reads the source text, detected domain, relevant terms, and TM examples via `get_prompt_context`, then drafts and saves the prompt. How much source it sees is set under Settings → AI Settings → “Prompt context – source segments”; 0 = the whole document.)* Built-in default prompts are protected – the AI saves your version under a new name rather than overwriting them. ### Working across sources Because the AI has all tools in one conversation, the most powerful prompts combine them: * “Compare how I translated *closing force* in this project vs my TM – if they differ, tell me which is more common and align the project.” * “Draft the remaining segments, but first build a glossary from the segments I already translated and stick to it.” * “Review my Draft segments against the source: flag mistranslations, fix typos directly, and list anything you weren’t sure about.” Given a long enough task, an assistant will keep going on its own:  **ChatGPT desktop.** Asked to carry on unattended, reporting back on a 1,064-segment file it translated, proofread and saved through the MCP server. Everything it did landed in Trados Studio, where it can be reviewed segment by segment like any other work. Worth knowing before trying it: the assistant writes into your project as it goes, so review the result as you would a colleague’s draft – the confirmation status it leaves, and Trados’s own verification, are what make that practical. An unattended run also spends tokens without you watching, so set a budget you are comfortable with first. Version tags like *(from v18.20.111)* show the plugin version a capability first shipped in – if the AI doesn’t offer it, update the plugin. New tools appear in your AI app automatically after a plugin update; no extension reinstall is needed. ## Setting it up  The External AI assistants (MCP) section at the bottom of the AI Settings tab. 1. In Trados Studio, open **Supervertaler Settings → AI Settings** and click **Connect AI assistant…** at the bottom. The dialog shows your current connection status. 2. **Claude Desktop** (easiest): click **Download extension (.mcpb)** to get `Supervertaler-MCP-Server.mcpb`. Then in Claude Desktop open **Settings → Extensions** and **drag the `.mcpb` file onto the page** – it shows a *“Drag .MCPB or .DXT files here to install”* target. (Prefer a file picker? Scroll to **Advanced settings** and use the **Install extension…** button instead.) Confirm the install. Double-clicking the `.mcpb` only works if your system has associated that file type with Claude Desktop; many don’t and will ask which app to use – just cancel and drag-and-drop instead. 3. **ChatGPT desktop** *(Windows)*: ChatGPT desktop can run the server, but it has no drag-and-drop installer, so this is a short manual step. It takes about two minutes. **a. Get the server.** In the **Connect AI assistant…** dialog click **Download server (.zip)** (or take `Supervertaler-MCP-Server-exe.zip` from any [GitHub release](https://github.com/Supervertaler/Supervertaler-for-Trados/releases/latest)). Unzip it and move `SupervertalerMcpServer.exe` somewhere **permanent** – for example `C:\Users\\Supervertaler\mcp\`. Do not leave it in Downloads: the path goes into a config file, and moving or clearing the file later breaks the connection. **b. Open the config file.** ChatGPT desktop reads its MCP servers from the same file as Codex CLI: ```plaintext %UserProfile%\.codex\config.toml ``` Paste that path into File Explorer’s address bar to jump straight to the folder. If the file or the `.codex` folder does not exist yet, create them – a config with only the block below in it is perfectly valid. **c. Add the server.** Append this to the end of the file, replacing the path with where you actually put the exe: ```toml [mcp_servers.supervertaler] type = "stdio" command = 'C:\Users\\Supervertaler\mcp\SupervertalerMcpServer.exe' args = [] enabled = true ``` Use **single quotes** around the path, as shown. In TOML that makes it a literal string, so Windows backslashes are taken exactly as typed. With double quotes you would have to write every backslash twice. **d. Restart ChatGPT desktop properly.** Closing the window is not enough – it keeps running in the notification area. Right-click its icon there and quit, then start it again. **e. Check it.** With Trados Studio running, ask ChatGPT: *“What Trados project is open?”* It should name your project, and `SupervertalerMcpServer` should appear under **Sources** in the reply. > **If nothing happens**, work through these in order: is Trados Studio actually running? Does the path in `config.toml` point at a file that exists? Did you fully quit ChatGPT from the notification area? And is the rest of the file still valid TOML – a stray character anywhere in it can stop *every* server loading, not just this one. 4. **Mistral Vibe CLI** *(Windows)*: Vibe is Mistral’s terminal coding agent – the CLI half of the product that used to be called le Chat. It runs local STDIO servers, so it can drive Trados just as Claude Code does. Worth knowing about if you need an EU-based, GDPR-compliant provider, or want to work on Mistral’s free tier. You chat in a terminal window rather than a polished desktop app; there is no Mistral desktop client, and Vibe’s *web* connectors are remote-only, so the CLI (or Vibe inside VS Code, JetBrains or Zed) is the only route to a local server. **a. Get the server.** Exactly as in step 3a above – **Download server (.zip)**, unzip it, and put `SupervertalerMcpServer.exe` somewhere permanent. **b. Add it to `config.toml`.** Vibe keeps its settings in: ```plaintext %UserProfile%\.vibe\config.toml ``` Append this, replacing the path with your own: ```toml [[mcp_servers]] name = "supervertaler" transport = "stdio" command = ['C:\Users\\Supervertaler\mcp\SupervertalerMcpServer.exe'] args = [] startup_timeout_sec = 30.0 tool_timeout_sec = 300.0 ``` Note the shape of `command`: **square brackets around a single-quoted path**. This one detail is what most Windows setups get wrong, and it fails in a confusing way – see the box below. Note also the double square brackets on `[[mcp_servers]]`; Vibe’s server list is a TOML array of tables, not a single table like ChatGPT’s. > **Why the brackets matter.** If you give `command` as a plain string, Vibe splits it with POSIX shell rules, which treat `\` as an escape character – so `C:\Users\you\...` silently becomes `C:Usersyou...`, a path that doesn’t exist. The server then never starts, and Vibe shows a connection error that does not go away. Wrapping the path in `[ ]` makes it a ready-made argument list that skips the splitting entirely. Writing the path with forward slashes (`C:/Users/you/...`) works too, and Windows accepts it. > > The same applies to `vibe mcp add … --command …`, which stores the path as a string: on Windows, prefer editing `config.toml` by hand. **c. Why the two timeouts.** Vibe defaults to a 10-second server start-up limit and a 60-second cap per tool call. Trados can take longer than that to answer while a project is loading, and the long-running tools (pre-translation, verification, comparing a document against a TM) routinely exceed a minute. The values above give them room; the bridge’s own limit is 5 minutes. **d. Check it.** Start Trados Studio **first**, then run `vibe` and type `/mcp` (or `/connectors`). Your server should be listed; `/mcp supervertaler` lists the tools it exposes. Then ask: *“What Trados project is open?”* 5. **Google Antigravity** *(confirmed September 2026)*: Antigravity is Google’s agentic desktop app – the Google-side counterpart to Claude Desktop and ChatGPT desktop, and the route to driving Trados from Gemini with a real interface rather than a terminal. > **Not Gemini Code Assist in VS Code.** Google has retired the free Gemini Code Assist tier for individuals in the VS Code extension. Signing in now returns *“This client is no longer supported for Gemini Code Assist for individuals”* and points you at Antigravity. Don’t spend time on that route. **a. Get Antigravity.** Download it from [antigravity.google/download](https://antigravity.google/download) and sign in with a **personal Google account**. A paid Google **Workspace** account – your own domain – is refused: Google restricts the free tier to consumer accounts, and a Workspace sign-in fails with a licence error. (During onboarding you’re offered “Build with Google” plugins for Firebase, Flutter, Maps and so on. None are relevant to translation work; leave them all unticked, as each adds its own tools alongside Supervertaler’s.) **b. Get the server.** Exactly as in step 3a – **Download server (.zip)**, unzip it, and put `SupervertalerMcpServer.exe` somewhere permanent. **c. Add the server.** Open **Settings → Customizations**, scroll to **Installed MCP Servers**, and click **Add MCP**. Or edit the config file directly – **Open MCP Config** in that same panel takes you to it: ```plaintext %UserProfile%\.gemini\config\mcp_config.json ``` ```json { "mcpServers": { "supervertaler": { "command": "C:\\Users\\\\Supervertaler\\mcp\\SupervertalerMcpServer.exe", "args": [] } } } ``` JSON requires every backslash to be doubled, as shown – or write the path with forward slashes, which Windows also accepts. Quit Antigravity completely afterwards and start it again; the file is read at startup. **d. Check it.** **Settings → Customizations → Installed MCP Servers** should list `supervertaler` with a green dot and a tool count – *“51 tools enabled”* at the time of writing, more as the plugin gains tools. Expand it to see the tool names. Then, with Trados Studio running, ask: *“What Trados project is open?”* > **Don’t test it by asking the AI to list its MCP tools.** It will tell you, confidently, that it has none – even with every tool connected and working. Models are unreliable about their own toolset. Trust the Settings panel and a real question instead. **e. Teach it to use the tools.** This step matters more here than with any other client. Antigravity’s agent has a shell and a file browser, and left to itself it will try to answer Trados questions by digging through `projects.xml` and your project folder – slowly, and from stale data, because neither reflects what Studio currently has open. On one 675-segment project that meant eight minutes, thirteen shell commands and no answer; the same question with the tools named took fourteen seconds. Supervertaler ships the instruction file that fixes this. In Trados, open **Supervertaler Settings → AI Settings → Connect AI assistant…**, scroll to **Google Antigravity**, and click **Install Antigravity skill**. It writes: ```plaintext %UserProfile%\.gemini\config\skills\supervertaler-trados\SKILL.md ``` Then restart Antigravity and check **Settings → Customizations** – it appears as `supervertaler-trados`, tagged `Global`. Start a **new conversation**; one already running keeps the instructions it began with. The Gemini CLI reads the same folder, so it is covered by the same click. The skill tells the agent to load your recorded decisions first, then run the `check_*` tools, then read the bilingual text – and not to write scripts to do any of it. Because it ships with the plugin it stays current: it names specific tools and rules, and a copy pasted by hand would slowly start describing a toolset that no longer matches. If you have edited your own copy, the button backs it up before replacing it. > **Gemini CLI** (terminal, no interface) reads a *different* file. Register the server with `gemini mcp add supervertaler --scope user "C:\Users\\Supervertaler\mcp\SupervertalerMcpServer.exe"`, which writes to `%UserProfile%\.gemini\settings.json`. Watch for its folder-trust gate: in a folder you haven’t trusted it disables every MCP server – including user-scope ones – and says so when you run `gemini mcp list`. 6. **Other MCP clients (Claude Code, etc.)**: click **Copy config snippet** and paste it into the app’s MCP configuration, adjusting the path to where you saved `SupervertalerMcpServer.exe`. The snippet is in Claude’s JSON format; clients that use a different format need the same two facts – the transport is STDIO, and the command is the path to that exe. Then open a project document in the Trados editor, and ask your AI app: *“What’s the status of my Trados project?”* > **Tip.** The connection starts automatically **as soon as Trados Studio is running** – no document or panel needed, so machine-wide questions (“what projects do I have?”) work straight from the Projects view. (History: on 18.20.99–18.20.111 the connection started when you opened a document in the editor; before 18.20.99 you had to click the Supervertaler Assistant panel once per session.) Install the extension **or** use a manual config entry – not both, or every tool will appear twice in the AI app. The Connect dialog warns you if it detects this. ## Privacy and security Everything stays on your computer: * The connection between the AI app and Trados runs over **localhost only** – nothing is exposed to your network or the internet. * Every Trados session uses a **fresh access token**; only programs on your own machine that hold the token can connect. * Your project data goes to an AI model only when *you* ask the AI a question about it, through the AI app you chose – exactly as if you had pasted the text yourself. The MCP server itself sends nothing anywhere. ## Requirements * Supervertaler for Trados with an active licence or trial (the bridge is part of the AI Assistant). * An MCP client that runs local STDIO servers on your own machine: Claude Desktop (recommended, one-click install), ChatGPT desktop, Google Antigravity, Claude Code, Gemini CLI, Mistral Vibe CLI, or similar. This means an app that executes the server **on your PC** – the claude.ai and chatgpt.com *websites*, and Mistral’s Vibe web app, all run servers in the cloud and cannot reach a local one. * Windows (the MCP server is a self-contained exe; no additional runtimes needed). ## Keeping it up to date **The MCP server is a separate component and does not update with the plugin.** Updating Supervertaler for Trados through the App Store updates the half that lives inside Studio; the `.mcpb` extension (or the unzipped exe) is installed in your AI app and stays exactly as it was until you replace it. The two halves talking to each other across a version gap is the single most common source of odd behaviour. After a plugin update that mentions the MCP server in its release notes, reinstall the extension: download the current `.mcpb` from the [latest release](https://github.com/Supervertaler/Supervertaler-for-Trados/releases/latest), install it the same way you did the first time, and restart your AI app. Installing over the existing extension is enough; there is nothing to uninstall first. Good news first: **the list of tools is not baked into it.** The server asks Studio for the current tool list every time it starts, so tools and options added by a plugin update work with an older server – provided Studio was running when your AI app started. What *is* fixed at install time is the server’s own behaviour, and the one symptom worth recognising is: * **Errors mentioning a timeout of 30 seconds.** That limit was raised to 5 minutes in v18.20.148, so if you still see it, the installed server predates that release and should be replaced. ## Troubleshooting * **The AI doesn’t know about a tool or option the release notes describe** – this is almost never a stale download. The tool list is read from Studio once, when your AI app starts, and kept for that whole session, so an option added by a plugin update is missing simply because the AI app was already running (or started while Studio was closed) – a new option is then dropped from the call silently rather than reported as an error. The fix is order, not reinstallation: **start Trados Studio first, then start (or fully restart) your AI app.** * **Double-clicking the `.mcpb` file asks which app to open it with** – your system has no `.mcpb` association. Cancel the dialog and instead either **drag the `.mcpb` onto the Extensions page**, or use Claude Desktop’s **Settings → Extensions → Advanced settings → Install extension…** button. (Drag-and-drop works once the Extensions page has finished loading – if it’s stuck on “Loading extensions…”, see the next point first.) * **The Extensions page is stuck on “Loading extensions…”** – the page needs to reach Anthropic’s extension directory once before it renders; we’ve seen it hang on the Microsoft Store build of Claude Desktop. Fully quit Claude Desktop (including the system tray icon) and reopen it; check your internet connection. If it keeps hanging, there’s a universal fallback that skips the Extensions page entirely: download `Supervertaler-MCP-Server-exe.zip` instead, unzip it somewhere permanent, and use the **Copy config snippet** button in the plugin’s Connect dialog to add the server manually to `claude_desktop_config.json` (Claude Desktop → Settings → Developer → Edit Config). * **The AI says it can’t reach Trados** – make sure Trados Studio is running; from v18.20.112 the connection starts with Studio itself (on 18.20.99–18.20.111 you additionally needed a document open in the editor, and before that a click on the Supervertaler Assistant panel – updating the plugin removes those steps). The Connect dialog’s status lines show whether the connection is up. Tools that read the open document still need one open, and will say so. * **Tools appear twice in Claude Desktop** – you have both the extension and a manual config entry; remove one (see above). * **Mistral Vibe CLI shows a permanent “cannot connect” message, but `/mcp` lists the server as enabled** – `/mcp` reports what is *configured*, not what is *connected*, so the two are not in conflict: the banner is right and the server really did fail to start. On Windows the usual cause is the path in `command` being written as a plain string, which Vibe splits with POSIX rules and strips the backslashes from. Put the path in square brackets, or use forward slashes – see step 4 of [Setting it up](#setting-it-up). A second, milder cause is Vibe’s 10-second start-up limit expiring while Studio is still busy; raise `startup_timeout_sec`. * **The AI refuses to change segments and mentions two instances** – you have both Studio 2024 and Studio 2026 open, and it will not guess which one you mean. Tell it (*“use the 2026 one”*), or close the Studio you are not working in. Reading still works throughout. See [Two Studios open at once](#two-studios-open-at-once-from-v1820184). * **Term lookups return nothing** – check that your termbase/database path is set correctly in the Supervertaler settings (the same path TermLens uses). * The bridge writes a diagnostic log to `\trados\runtime\bridge.log`. Development of this feature is tracked publicly in [issue #44](https://github.com/Supervertaler/Supervertaler-for-Trados/issues/44) – feedback and use-case ideas are very welcome. # MultiTerm Support TermLens automatically detects MultiTerm termbases (`.sdltb` files) attached to your active Trados project and displays their terms alongside your Supervertaler terms. Note This page covers **Trados Studio 2024**, which uses MultiTerm `.sdltb` termbases. If you are on **Trados Studio 2026**, terminology comes from the new `.ttb` format instead –see [Trados Studio 2026 & .ttb](/trados/studio-2026/). ### How It Works When you open a project in Trados Studio that has MultiTerm termbases attached, TermLens reads those `.sdltb` files and loads all term pairs into its matching engine. MultiTerm terms appear as **green chips** in the TermLens panel, right next to the blue, pink, and yellow chips from your Supervertaler termbases. There is nothing to configure. If your Trados project has MultiTerm termbases attached and **enabled** (via **Project Settings > Language Pairs > Termbases**), TermLens picks them up automatically. Termbases with the **Enabled** checkbox unchecked in Trados are ignored. #### Colour coding | Colour | Meaning | | ------------ | ---------------------------------------------------- | | **Blue** | Regular Supervertaler termbase match | | **Pink** | Project termbase match (higher priority) | | **Yellow** | Non-translatable term (source = target) | | **Green** | MultiTerm termbase match (`.sdltb`) | | **Lavender** | Abbreviation match (matched via source abbreviation) | Green chips behave like any other TermLens chip – click to insert, or use **Alt+1** through **Alt+9** to insert by number. ![]() ### Read-Only MultiTerm termbases are **read-only** in TermLens. You cannot add, edit, or delete terms in a MultiTerm termbase from the TermLens panel. To manage MultiTerm terms, use Trados Studio’s built-in MultiTerm interface. When you right-click a green MultiTerm chip, the Edit, Delete, and “Mark as Non-Translatable” options are not shown. ### Auto-Refresh TermLens monitors your MultiTerm termbases for changes: * **Term changes** –when you add or edit terms using Trados’s native MultiTerm interface, TermLens detects the file change on the next segment navigation and reloads automatically. * **Config changes** –when you enable or disable a MultiTerm termbase in **Project Settings > Termbases**, TermLens detects the change within a few seconds and updates the panel automatically – no segment change needed. ### MultiTerm Termbases in Settings MultiTerm termbases appear at the bottom of the termbase list in the **Supervertaler Settings** dialogue (gear icon > TermLens tab). Each one is labelled with **\[MultiTerm]** and has a light green background to distinguish it from Supervertaler termbases. | Toggle | Behaviour | | ----------- | ------------------------------------------------------------------------------ | | **Read** | Controls whether this termbase’s terms appear in TermLens. Uncheck to hide it. | | **Write** | Always disabled –MultiTerm termbases are read-only in TermLens | | **Project** | Always disabled –only Supervertaler termbases can be the project termbase | To add or remove MultiTerm termbases from your project, use Trados Studio’s **Project Settings > Language Pairs > Termbases**. ### MultiTerm and AI Terminology Injection Your MultiTerm termbases are loaded and used for **AI terminology injection**. This means the AI Assistant, Batch Translate, and Ctrl+T all receive your MultiTerm terminology in their prompts, helping the AI use the correct approved terms – as long as the termbases are enabled in Trados Project Settings. ### Technical Details TermLens reads `.sdltb` files directly using the JET 4.0 database driver built into Windows. This is the same driver that MultiTerm itself uses. If the JET driver is not available (uncommon on modern Windows), TermLens falls back to Trados’s terminology provider API for per-segment lookups. Because the access is read-only, there is no risk of data corruption. TermLens opens the `.sdltb` file in shared read mode, so MultiTerm and Trados can continue to use it simultaneously. ### Troubleshooting #### MultiTerm terms not appearing 1. **Check the Trados Enabled checkbox** –open **Project Settings > Language Pairs > Termbases** and make sure the termbase’s **Enabled** checkbox is ticked 2. **Check the Read toggle** –open Supervertaler Settings and make sure the MultiTerm termbase’s Read checkbox is enabled 3. **Check languages** –the termbase’s source and target languages must match the current project’s language pair #### Terms added in MultiTerm not updating * Navigate to a different segment –this triggers the auto-refresh check * If terms still do not appear, close and reopen the settings dialogue to force a full termbase reload *** ### See Also * [TermLens](/trados/termlens/) * [Termbase Management](/trados/termbase-management/) * [TermLens Settings](/trados/settings/termlens/) * [Troubleshooting](/trados/troubleshooting/) # Privacy For the full Supervertaler privacy policy, please visit: **[supervertaler.com/privacy](https://supervertaler.com/privacy/)** ## Summary * Supervertaler does **not** collect, store, or transmit your translations, termbases, or documents to Supervertaler servers * Your translations, termbases, and documents stay on your machine * When you use AI features, data is sent directly to the AI provider **you** selected, using your own API key * **Optional anonymous usage statistics** – strictly opt-in. If you consent, a single ping on startup sends only: plugin version, OS version, Trados version, and system locale. No personal data, no translation content. You can opt out at any time in Settings. See [Usage Statistics](/trados/settings/usage-statistics/) for full details. * **Trial registration** (v18/19.20.118+) – trial installs register the trial’s start date with a first-party licence endpoint on startup, so the trial behaves consistently across reinstalls. Only the anonymous machine hash, plugin/Studio version, locale, and the trial’s start date are sent – no content, no personal data. This is part of how the trial functions (like licence validation) and is separate from the opt-in usage statistics. See the [full privacy policy](https://supervertaler.com/privacy/) for details. * For maximum confidentiality with AI features, use [Ollama](https://ollama.com) – all processing stays local * Source code is publicly available on [GitHub](https://github.com/Supervertaler/Supervertaler-for-Trados) for independent verification ## Contact Questions about privacy? Email . # QuickLauncher QuickLauncher gives you one-click access to your most-used AI prompts directly from the Trados editor, without switching panels or typing anything. ### How it works 1. **Right-click** anywhere in the editor (or press `Alt+Q`) 2. Click **QuickLauncher** in the context menu 3. Select a prompt from the list 4. The prompt is filled in with the current segment context and submitted to the Supervertaler Assistant chat  The QuickLauncher context menu with folder sections and keyboard shortcuts. Note The menu heading **Supervertaler QuickLauncher** is clickable – click it to open **Settings → Prompts**, where you can view, edit, and organise your QuickLauncher prompts. The expanded prompt appears as a user message bubble in the **Supervertaler** chat panel, and the AI response follows immediately below it. The conversation continues from there – you can ask follow-up questions in the chat input as normal. ### QuickLauncher vs Translate Active Segment (Alt+T) These are two different pipelines, and the difference matters: * **Translate active segment (`Alt+T`)** runs through the **Batch Translate pipeline**: it uses the prompt currently selected in the Batch Translate tab, plus the same provider, termbase terms, document context and SuperMemory context as a batch run, and writes the result straight into the target cell. *(Document and SuperMemory context from v18/19.20.149 – before that, single segments were sent without context, which made them translate noticeably worse than a batch.)* * **QuickLauncher** runs through the **Assistant chat**: it fills in whichever prompt template you assigned to the slot and shows the answer in the chat panel. What context the AI sees depends entirely on the template – surrounding text only if it contains `{{SURROUNDING_SEGMENTS}}`, the whole document only if it contains `{{PROJECT}}`, and so on (see the variables below). If your goal is simply *“translate this segment the way a batch would”*, use `Alt+T`. Use QuickLauncher when you want a custom prompt’s behaviour – explaining a term, assessing a translation, or translating with special instructions of your own. ### Keyboard shortcut | Shortcut (Windows) | Shortcut (Mac) | Action | | ------------------ | -------------- | ------------------------------ | | `Alt+Q` | `Option+Q` | Open QuickLauncher prompt menu | Note **Changed in v18/19.20.184: this used to be `Ctrl+Q`.** That collided with Trados’s own **View Internally Source**, and Trados wins – so QuickLauncher silently did nothing until you cleared that binding yourself. `Alt+Q` needs one small setup step: Trados also assigns it to **Tell me what you want to do** (confirmed in Studio 2024), and Trados wins, so clear that binding in **File → Options → Keyboard Shortcuts** – Tell Me is the ribbon search box at the top right and works fine without a shortcut. If you had already cleared the Trados binding, `Ctrl+Q` will now do nothing for QuickLauncher; use `Alt+Q` instead, or reassign it under **File → Options → Keyboard Shortcuts**. ### Prompt variables QuickLauncher prompts have access to the full segment and project context at the moment you trigger them. #### Language variables | Variable | Replaced with | Example | | --------------------- | -------------------------------------- | ------------------------- | | `{{SOURCE_LANGUAGE}}` | Source language name, including locale | `Dutch (Belgium)` | | `{{TARGET_LANGUAGE}}` | Target language name, including locale | `English (United States)` | #### Segment variables | Variable | Replaced with | Example | | -------------------- | -------------------------------------------------------------------- | -------------------------------------- | | `{{SOURCE_SEGMENT}}` | Full text of the **active source segment** | `De uitvinding heeft betrekking op...` | | `{{TARGET_SEGMENT}}` | Full text of the **active target segment** (your translation so far) | `The invention relates to...` | | `{{SELECTION}}` | Text currently **selected** in the editor | `werkwijze` | Note **Segment vs selection:** `{{SOURCE_SEGMENT}}` and `{{TARGET_SEGMENT}}` always give the **entire active segment**. `{{SELECTION}}` gives only the **highlighted portion** – useful for term lookups or focused questions. If nothing is selected, `{{SELECTION}}` is an empty string. #### Project variables | Variable | Replaced with | | -------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------- | | `{{PROJECT_NAME}}` | Trados project name (e.g. `Patent_NL_EN_2026`) | | `{{DOCUMENT_NAME}}` | Active file name (e.g. `source_document.docx`) | | `{{SURROUNDING_SEGMENTS}}` | N source segments before and after the active segment, with their actual Trados segment numbers and the active segment marked `← ACTIVE` | | `{{PROJECT}}` | All source segments in the active document, numbered with their actual Trados segment numbers | | `{{TM_MATCHES}}` | Translation memory fuzzy matches (≥70%) for the active segment, showing match percentage, source, and target text | **`{{SURROUNDING_SEGMENTS}}` example output** (with N = 2): ```plaintext [11] Vorige zin hier. [12] Nog een vorige zin. [13 ← ACTIVE] De uitvinding heeft betrekking op een nieuwe werkwijze... [14] Volgende zin hier. [15] Nog een volgende zin. ``` **`{{PROJECT}}` example output** (single-file project): ```plaintext [1] De uitvinding heeft betrekking op een nieuwe werkwijze... [2] De conclusies omvatten de volgende kenmerken... [3] ... ``` In a **multi-file project**, a file header is inserted at each boundary (because Trados restarts segment numbering per file): ```plaintext === File 1 === [1] Conclusie 1 omvat... [2] Conclusie 2 omvat... === File 2 === [1] De beschrijving begint hier... ``` Caution `{{PROJECT}}` sends all source segments to the AI. For a typical 10,000-word patent this costs roughly **4–5 cents** per call with a Sonnet-class model – negligible for important work, but avoid using it in high-frequency prompts. The number of surrounding segments for `{{SURROUNDING_SEGMENTS}}` is configured in **Settings → AI Settings → Surrounding segments** (default: 5). To keep the chat history readable, the chat bubble shows a compact summary (e.g. `[source document – 47 segments]`) instead of the full source text. The complete document is still sent to the AI. #### Example: explain a selected term Select a word in the source segment, press `Alt+Q`, and choose a prompt like this: ```plaintext The user is translating from {{SOURCE_LANGUAGE}} to {{TARGET_LANGUAGE}}. The selected term is: {{SELECTION}} Explain what this term means and suggest the best {{TARGET_LANGUAGE}} equivalent, considering the full segment context below: {{SOURCE_SEGMENT}} ``` #### Example: assess the current translation ```plaintext Source ({{SOURCE_LANGUAGE}}): {{SOURCE_SEGMENT}} My translation ({{TARGET_LANGUAGE}}): {{TARGET_SEGMENT}} Assess how I translated the current segment. Point out any inaccuracies, awkward phrasing, or terminology issues, and suggest improvements. ``` #### Example: translate a selected term using surrounding context Uses `{{SELECTION}}` together with `{{SURROUNDING_SEGMENTS}}` so the AI sees the passage around the active segment, not just the active segment itself: ```plaintext I am translating a {{SOURCE_LANGUAGE}} patent into {{TARGET_LANGUAGE}}. The selected term is: {{SELECTION}} Here is the passage surrounding the active segment for context: {{SURROUNDING_SEGMENTS}} Suggest the best {{TARGET_LANGUAGE}} translation for "{{SELECTION}}" given the surrounding context. Give a brief explanation of your reasoning. ``` #### Example: full-document term query Uses `{{PROJECT}}` to give the AI the complete source text. Reserve this for high-stakes queries where full document context matters, such as a key term that appears in multiple places with different nuances: ```plaintext I am translating a {{SOURCE_LANGUAGE}} patent ({{DOCUMENT_NAME}}) into {{TARGET_LANGUAGE}}. Project: {{PROJECT_NAME}} Here is the complete source text, segment by segment: {{PROJECT}} Throughout this document, what is the most accurate and consistent {{TARGET_LANGUAGE}} translation for "{{SELECTION}}"? Consider all occurrences in context and note any variation in meaning between them. ``` #### Example: check specific segments by number After using `{{PROJECT}}` the AI knows the segment numbers, so you can follow up in the chat – or build a prompt that asks about specific segments from the start: ```plaintext I am translating a {{SOURCE_LANGUAGE}} patent into {{TARGET_LANGUAGE}}. Here is the source document: {{PROJECT}} My translations so far: - Segment 1: [paste your translation here] - Segment 4: [paste your translation here] Do you think these translations are accurate and consistent with the terminology used elsewhere in the document? ``` #### Example: translate using TM fuzzy matches Uses `{{TM_MATCHES}}` to give the AI any high fuzzy matches from the translation memory, so it can leverage existing translations as a starting point: ```plaintext Translate the following segment from {{SOURCE_LANGUAGE}} to {{TARGET_LANGUAGE}}. Source: {{SOURCE_SEGMENT}} Here are fuzzy matches from my translation memory: {{TM_MATCHES}} Use the fuzzy matches as reference where helpful, but produce an accurate translation of the source segment – do not simply copy a fuzzy match. ``` Note `{{TM_MATCHES}}` only includes matches of **70% or higher**. If no matches meet this threshold, the variable is replaced with “(no fuzzy matches above 70%)”. The match data comes from the active segment’s translation origin in Trados – the same match shown in the Translation Results pane. The plugin fills in all variables and sends the expanded prompt straight to the AI. ### Folder display mode By default, subfolders in the QuickLauncher menu appear as **expandable submenus** (hover to open). You can change any folder to display as a **flat section** instead – its prompts appear directly in the main menu under a bold header, with separators between sections. To toggle the display mode: 1. Open **Settings → Prompts** 2. Right-click a QuickLauncher folder in the tree 3. Click **Show as section in menu** (a checkmark indicates the current state) This setting is per-folder, so you can mix styles – for example, keep a large folder as an expandable submenu while showing a small one as a flat section. ### Setting up QuickLauncher prompts Set `category: QuickLauncher` in the YAML frontmatter, or place the file in a folder called `QuickLauncher` inside your `prompt_library`. See [Prompts → Marking a prompt as a QuickLauncher shortcut](/trados/settings/prompts/#marking-a-prompt-as-a-quicklauncher-shortcut) for full details. ### Shared with Supervertaler for memoQ QuickLauncher prompts live in the shared `prompt_library` folder used by both Supervertaler for Trados and Supervertaler for memoQ. Any prompt you create in one application is immediately available in the other. #### YAML reference If you prefer editing prompt files directly, the relevant frontmatter fields are: ```yaml --- name: Explain selected term category: QuickLauncher quicklauncher_modes: [assistant, clipboard] default_mode: assistant --- ``` * `quicklauncher_modes:` – accepts an inline list `[assistant, clipboard]` or a comma-separated string `assistant, clipboard`. Unknown values are silently dropped. When omitted, defaults to `[assistant]`. * `default_mode:` – which mode appears first in the submenu when both are configured. Must be one of the values in `quicklauncher_modes`. Defaults to `assistant`. *** ### See Also * [Text Transforms](/trados/text-transforms/) * [Prompts](/trados/settings/prompts/) * [Supervertaler](/trados/ai-assistant/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) # Reports The **Reports** tab in the Supervertaler Assistant panel is where the assistant collects structured output from its AI operations. Two kinds of thing land here: **proofreading results** and an optional **log of AI calls**. ## Proofreading results When you run the **AI Proofreader** (the Proofread mode of [Batch Operations](/trados/batch-operations/)), each issue the AI finds is shown here as a clickable card. See [AI Proofreader](/trados/ai-proofreader/) for the full workflow and what each card contains. Since v18.20.187 every completed run is also written to `trados\ eports` in your data folder as Markdown, and **Save report…** next to **Clear** writes a copy wherever you choose. The button is enabled only while a proofreading report is showing. ## AI operation log When **Log prompts and responses to Reports tab** is enabled in [AI Settings](/trados/settings/ai-settings/), AI calls – Chat, Batch Translate, Batch Proofread and AutoPrompt – are recorded here together with the prompt, the response, the model used and the token/cost figures. It is the place to audit exactly what was sent to the AI provider and to review cost after the fact. # Ai Settings Configure the AI provider, model, and context options used by the Supervertaler for Trados plugin. ## Accessing AI settings Open the plugin **Settings** dialogue and switch to the **AI** tab. ## Provider selection Choose one of the supported AI providers: | Provider | Models | Where to get a key | | ------------------------------ | ------------------------------------------------------------------ | -------------------------------------------------------------------- | | **OpenAI** | GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.4 Mini | [platform.openai.com/api-keys](https://platform.openai.com/api-keys) | | **Claude (Anthropic)** | Claude Sonnet 5, Claude Haiku 4.5, Claude Opus 5, Claude Fable 5.1 | [console.anthropic.com](https://console.anthropic.com) | | **Gemini (Google)** | Gemini 3.8 Flash, Gemini 3.5 Flash-Lite, Gemini 3.1 Pro (Preview) | [aistudio.google.com/apikey](https://aistudio.google.com/apikey) | | **Grok (xAI)** | Grok 4.3 | [console.x.ai](https://console.x.ai) | | **Mistral AI** | Mistral Large, Mistral Small | [console.mistral.ai](https://console.mistral.ai) | | **DeepSeek** | DeepSeek V4 Pro, DeepSeek V4 Flash | [platform.deepseek.com](https://platform.deepseek.com) | | **[OpenRouter](#openrouter)** | 200+ models from all major providers, one key | [openrouter.ai/keys](https://openrouter.ai/keys) | | **Ollama (Local)** | Runs models on your own machine | No key needed | | **Custom (OpenAI-compatible)** | Any OpenAI-compatible API | From whoever runs the endpoint | Note You only need one provider to get started. If you would rather not manage several accounts, **OpenRouter** gives you access to models from all of the above with a single key. ### Getting your first key 1. Pick a provider from the table above and follow its link. 2. Create an account and generate an API key. Most providers require billing details before a key will return anything; the free tiers that do exist are usually rate-limited rather than free-forever. 3. Copy the key **immediately** – most providers show it only once. 4. Paste it into the **API key** field below, then use **Test Connection** to confirm it works before you rely on it mid-job. Keys are stored locally in your Supervertaler data folder and are only ever sent to the provider’s own API endpoint. See [Privacy](/trados/privacy/) for what does and does not leave your machine. Caution An API key is a billing credential. Anyone who has it can spend money on your account. Do not paste one into a shared prompt, a support ticket, or a screenshot – and if you think one has leaked, revoke it from the provider’s console rather than just replacing it here. ## API key Enter the provider’s API key here. **Test Connection** checks it with a small call. ### One key file for every product (from v18.20.187) Keys are kept in one file shared by Supervertaler for Trados, Supervertaler for memoQ and Supervertaler Sidekick: `settings\api-keys.json` in your Supervertaler data folder, one line per provider, in plain text. Whatever you type here is written to that file, and the other products read it, so a key pasted once works everywhere. The file is filled from the plugin’s existing keys the first time it is missing, so nothing needs retyping after the update. You can also edit it in Notepad. The key box says so when a key plainly belongs to another service. Each provider’s keys have a recognisable start: Anthropic `sk-ant-`, OpenAI `sk-` or `sk-proj-`, Gemini `AIza`, xAI `xai-`, OpenRouter `sk-or-`. ## Model selection A dropdown showing a curated list of recommended models for the selected provider. ### The short list, and all models (from v18.20.187) By default the dropdown is a short list: the few models worth recommending for translation work, each with a one-line verdict. It is kept current with every release, and a model that has been superseded is removed from it rather than left in with a warning. Tick **Show all models** beside the dropdown to see everything the provider offers as well. Click **Fetch list** to ask the provider for its current list, using the API key in the box (it need not be saved yet) – Anthropic, OpenAI, Gemini, Mistral, DeepSeek, xAI, OpenRouter, Ollama and custom OpenAI-compatible endpoints all publish one. Fetching switches **Show all models** on. Models not in the short list appear with the provider’s own name for them; dated snapshots and models that are not for translating text (speech, image, transcription, embeddings) are left out. The fetched list and the tick box are remembered across restarts, and the Batch Operations and chat model menus follow them. This is the way to use a model released after your plugin build: click **Fetch list**, pick it, OK. ### Model ID Below the dropdown is an optional **Model ID** field. To use a model that isn’t in the curated list – a brand-new release, a preview model, or an OpenRouter router such as `openrouter/free` – type its exact model ID here. When filled, it overrides the dropdown selection; leave it blank to use the model picked from the dropdown. The field is available for every cloud provider. If you reopen Settings and a saved model isn’t in the curated list, it is shown back in the Model ID field. ## Ollama endpoint When using Ollama as the provider, this field sets the local endpoint URL. Defaults to: ```plaintext http://localhost:11434 ``` Change this only if you are running Ollama on a different port or a remote machine. ## DeepSeek [DeepSeek](https://platform.deepseek.com) is a Chinese AI lab offering high-quality models with competitive pricing. To use DeepSeek directly: 1. Create an account at [platform.deepseek.com](https://platform.deepseek.com) 2. Go to **API Keys** and create a key 3. In Supervertaler, select **DeepSeek** as the provider and paste your key DeepSeek models are also available via [OpenRouter](#openrouter) if you prefer a single-key setup. ## Custom OpenAI-compatible provider For providers that expose an OpenAI-compatible API (e.g., Azure OpenAI, together.ai, internal LLM gateways, local inference servers), configure these fields: | Field | Description | | ------------ | ------------------------------------------------------------- | | **Endpoint** | The base URL for the API (e.g., `https://your-server.com/v1`) | | **Model** | The model identifier to use (e.g., `llama-3-70b`) | | **API Key** | The authentication key for this endpoint | ### Managing multiple endpoints You can configure more than one custom endpoint and switch between them without re-entering credentials. Each endpoint is stored as a named **profile** in the **Profile** dropdown. | Button | Action | | ------ | ----------------------------------------------------------------------- | | **+** | Add a new endpoint (starts as “New Endpoint 1”, “New Endpoint 2”, etc.) | | **−** | Remove the currently selected endpoint | | **✎** | Rename the currently selected endpoint | Names are free-form labels – use whatever makes sense for your workflow (e.g. `Azure Production`, `Internal gateway – Mistral Large`, `Local Ollama`). Names must be unique within the list. Renaming is a UI-only change: the endpoint URL, model, and API key all stay attached to the same profile. ## OpenRouter [OpenRouter](https://openrouter.ai) is an API gateway that gives you access to 200+ models from OpenAI, Anthropic, Google, Mistral, Meta, and many others – all through a single API key. Instead of managing separate keys for each provider, you sign up once at OpenRouter and use one key for everything. ### Getting started 1. Create a free account at [openrouter.ai](https://openrouter.ai) 2. Go to **Keys** and create an API key 3. In Supervertaler, select **OpenRouter** as the provider and paste your key ### Curated model list The model dropdown includes a curated selection of the best models for translation: | Model | Description | | ------------------------ | ------------------------------------------------------------ | | **Claude Sonnet 4.6** | Recommended – best balance of speed, quality, and cost | | **Claude Opus 4.8** | Highest quality – Anthropic’s most capable model, 1M context | | **GPT-5.5** | Premium quality – OpenAI’s most advanced model | | **GPT-5.4 Mini** | Fast, affordable, and high quality for everyday translation | | **Gemini 3.1 Pro** | Google’s most advanced model, large context | | **Gemini 3 Flash** | Fast and affordable – great for large batch jobs | | **Gemma 4 31B** | Open-source – strong multilingual quality, 256K context | | **Gemma 4 26B MoE** | Open-source – near-31B quality at a fraction of the cost | | **Mistral Small 4** | Very fast and cheap – good multilingual support | | **Qwen 3.6 Plus (Free)** | Free – no API costs, good general-purpose quality | | **DeepSeek V4 Pro** | DeepSeek flagship – strong multilingual, competitive pricing | | **DeepSeek V4 Flash** | DeepSeek fast – great for high-volume translation | ### Using any OpenRouter model OpenRouter exposes far more models than the curated list above. To use one that isn’t listed, type its exact model ID into the **Model ID** field (see [Model selection](#model-selection)) – for example, `meta-llama/llama-3.1-70b-instruct`, `deepseek/deepseek-r1`, or a router such as `openrouter/free`. Browse all available models at [openrouter.ai/models](https://openrouter.ai/models). ### Pricing OpenRouter adds a **5.5% platform fee** on top of the underlying provider’s token price. For example, if Claude Sonnet 4.6 costs $3/$15 per million tokens at Anthropic, it costs approximately $3.17/$15.83 through OpenRouter. For a typical 5,000-word translation costing $0.50, the OpenRouter fee adds less than 3 cents. Note OpenRouter also offers some **free models** (marked with “Free” in the dropdown). These have no API cost at all – they are rate-limited but perfectly usable for testing or light workloads. ## Context layer options These options control what additional context is included in AI prompts. The settings are split into two groups depending on which features they apply to. ### Which settings apply where | Setting | Chat & QuickLauncher | Batch Operations | | ------------------------------------ | :------------------: | :--------------: | | Termbases in AI prompts | Yes | Yes | | Include full document content | Yes | Yes | | Max segments | Yes | Yes | | Include term definitions and domains | Yes | Yes | | Log prompts to Reports | Yes | Yes | | Include TM matches | Yes | AutoPrompt only | | Surrounding segments | Yes | No | ### Context layers (Batch operations, Chat and QuickLauncher) These settings apply to **all** AI features – Chat, QuickLauncher, Batch Translate, and Batch Proofread. #### Include full document content When enabled, all source segments in the current document are sent to the AI so it can determine the document type (legal, medical, technical, marketing, etc.) and provide context-appropriate assistance. This uses more tokens but greatly improves response quality – the AI can tailor its terminology and style to the specific type of document you are translating. For very large documents, the content is automatically truncated to the configured maximum. The truncation preserves the beginning and end of the document (first 80% + last 20%). For Batch Operations, the document content is included once in the system prompt (shared across all batches), so the AI knows what kind of document it is translating even when processing individual batches of segments. #### Max segments The maximum number of source segments to include in the AI prompt when document content is enabled. Default: **500**. Range: 100–2000. Increase this for very large documents where you want the AI to see more content. Decrease it if you want to reduce token usage. Note This setting is only available when **Include full document content** is enabled. #### List numbering as structure context (from v18.20.188; always on from v18.20.189) Word numbers claims, letters steps and bullets lists as paragraph properties, not as text. The segment grid never contains the `a)` or the `9.`, and until v18.20.188 neither did anything the AI received. On a patent, that means the model reads six unlettered steps and can conclude the source forgot to letter them – on a real document it translated “steps a. to f.” faithfully and then flagged it as a possible source defect, a note that would have reached the client. Batch Translate, Translate Segment, Clipboard Mode and SuperBench prefix the first segment of every numbered paragraph with the marker Word renders, inside a sentinel: `[#e)]het fixeren…`, `[#9.]Werkwijze…`, `[#•]een behuizing…`. A rule in the plugin’s own preamble tells the model that this is structure – use it to resolve cross-references and keep list items parallel, never translate it, never include it in the output. The rule ships with every request, including prompts you wrote yourself and AutoPrompt’s, because it lives in the plugin and not in a prompt template. **Preview prompt** shows exactly what goes out. The markers come from the original Word file that Studio keeps inside the sdlxliff, and are computed for the whole document at once, so a list that restarts at claim 11 reads `11.` and lettered steps that continue from one claim into the next keep counting – exactly as Word shows them. The batch log says how many markers were found and how many segments in the run carry one. Two safety nets: anything the model echoes back in a target is removed before the segment is written and logged in the batch log, and the TMX backup records the segment as Studio has it, without the sentinel. Files that are not Word documents, and Word files without any lists, get a one-line fallback rule instead: the numbering exists, is not in the text, and its absence is not a defect. v18.20.188 shipped this behind a checkbox so the first real runs could be checked; they showed no marker reaching a target, so from v18.20.189 it is simply on and the checkbox is gone. If you ever need it off – say, to compare prompts – add `"structureContext": false` to the `aiSettings` block of `settings.json` in `%LocalAppData%\Supervertaler.Trados\` while Studio is closed. #### Include term definitions and domains When enabled, term definitions, domains, and usage notes from your termbases are included alongside matched terminology in the AI prompt. This gives the AI deeper understanding of your terminology – for example, knowing that a term belongs to the legal domain or has a specific definition helps the AI use it correctly in both chat responses and batch translations. #### Include termbases in AI prompt Select which termbases are included in AI prompts. Terminology matches from enabled termbases are injected into the prompt to help the AI use the correct, approved terminology. For [AutoPrompt](/trados/generate-prompt/), **TermScan** automatically filters the termbase to only terms that appear in the document’s source text, keeping the prompt focused and within token limits. Caution **Only enable termbases you trust.** The AI will follow your termbase entries even when they are wrong. If a termbase contains inaccurate, outdated, or low-quality translations, the AI will be forced to use them – producing worse results than if no termbase were enabled at all. Modern LLMs are remarkably good at choosing correct terminology on their own. When in doubt, disable termbases and add terms incrementally as you review the AI’s output. ### Context layers (mostly Chat and QuickLauncher) These settings apply primarily to the **Supervertaler** chat window and **QuickLauncher** prompts. The exception is **Include TM matches**, which also feeds AutoPrompt – see the per-setting notes below. #### Include TM matches The behaviour of this checkbox depends on which feature is asking for context: * **Chat and QuickLauncher (live TM lookups).** When enabled, the AI gets translation memory matches – fuzzy and exact – for the active segment. This gives the AI reference translations from your project TMs to improve consistency. * **AutoPrompt (Batch Operations).** When enabled, [AutoPrompt](/trados/generate-prompt/) samples up to 50 already-translated, human-confirmed segment pairs evenly from the active document and includes them in the meta-prompt as in-project reference translations. This includes 100% / exact matches that have been applied and confirmed, fuzzy-and-edited segments, and segments translated from scratch – any segment with a Translated, Approved, or Signed-off confirmation level qualifies. AutoPrompt does **not** do live TM lookups; it samples confirmed segments straight from the document. * **Other Batch Operations (Translate, Proofread).** Unaffected by this checkbox – they always work segment-by-segment without TM reference pairs, regardless of how it’s set. Tip **Tip for AutoPrompt users:** confirm a handful of segments you are happy with before clicking AutoPrompt. Even 10–20 confirmed segments give the AI meaningful style anchors to work from. Without any confirmed segments to sample, the generated prompt won’t have in-project reference translations. #### Surrounding segments The number of segments before and after the active segment to include as context. Default: **5** (five segments on each side). Range: 1–20. This provides the AI with local context around the segment you are working on. It is also used for the `{{SURROUNDING_SEGMENTS}}` variable in [QuickLauncher prompts](/trados/settings/prompts/prompt-variables/). Note Batch Operations do not use this setting because each batch already contains a group of segments that provide context for each other. Tip **Tip:** For the best results, enable all context options. The more information the AI has about your project, document, terminology, and previous translations, the more accurate and consistent its suggestions will be. ### SuperMemory context These two toggles control whether [SuperMemory](/trados/ai-assistant/super-memory/) content is included in the AI context. #### Include memory bank in AI context When enabled, the AI reads the active bank’s `brief.md`, `terminology.md` and `style.md` - plus the `_shared` bank of house defaults - before every translation and chat message. This gives it the *reasoning* behind your decisions, not just the terms themselves. Caution **Off by default.** SuperMemory is a power-user feature that most translators should opt into deliberately. The simpler workflow – TermLens termbases + the AI context options above – covers the majority of needs. Enable this toggle once you have a populated memory bank and want the AI to consult it. #### Use memory bank when generating prompts (AutoPrompt) When enabled, SuperMemory content is included in the [AutoPrompt](/trados/generate-prompt/) meta-prompt so that generated translation prompts reflect your established client conventions, terminology reasoning, and style guides. Only effective when “Include memory bank in AI context” is also enabled. ## Prompt logging ### Log prompts and responses to Reports tab When enabled, AI operations are logged to the **Reports** tab in the Supervertaler Assistant panel. Each log entry shows: * The **feature and prompt name** (e.g. “QuickLauncher · Explain in Context”) * The **model used**, estimated **token counts**, **cost**, and **duration** * Expandable sections for the **system prompt**, **messages**, and **response** Note **Batch Translate** operations appear as a single consolidated entry showing the combined token count, cost, and total duration for the entire operation – regardless of how many sub-batches were processed. Click “Show system prompt…”, “Show messages…”, or “Show response…” to expand a section. Press **Escape** to collapse it. Use **Copy** to copy a single section, or **Copy all** to copy the full prompt details to your clipboard. This is useful for: * **Monitoring costs** – see exactly how many tokens each operation uses * **Debugging prompts** – inspect the full text sent to the AI to understand its behaviour * **Comparing models** – run the same prompt with different models and compare token usage Note Prompt logging is off by default to keep the Reports tab clean. Enable it when you want to inspect or audit your AI usage. Log entries are stored in memory only and cleared when Trados restarts. ## Batch settings Configure the **batch size** for the [Batch Translate](/trados/batch-translate/) feature. This determines how many segments are sent to the AI provider in a single request. * A larger batch size is faster but uses more tokens per request * A smaller batch size is more granular and easier to review *** ## See Also * [Prompts](/trados/settings/prompts/) * [AI Cost Guide](/trados/ai-cost-guide/) * [TermLens Settings](/trados/settings/termlens/) * [Token Usage & Costs](/trados/usage-costs/) # Backup Use the **Export Settings** and **Import Settings** buttons in the **Backup** tab of the Settings dialogue to back up and restore your Supervertaler configuration. ## Export Click **Export Settings…** to save a copy of your current settings to a JSON file. Choose a location and filename – the default is `supervertaler-settings.json`. This file contains all your plugin settings: termbase paths, toggle states, font size, shortcut preferences, AI provider keys, model selections, and prompt configuration. Note **Tip:** Export your settings before upgrading the plugin or switching machines, so you can quickly restore your setup. ## Import Click **Import Settings…** to restore settings from a previously exported JSON file. The import process: 1. Validates that the selected file is a valid Supervertaler settings file 2. Creates an automatic backup of your current settings (`settings.backup.json`) 3. Replaces your current settings with the imported ones 4. Closes the Settings dialogue and applies the new settings immediately Caution Importing settings replaces **all** current settings. Your previous settings are automatically backed up in case you need to revert. ## Settings file location Your settings are stored at: ```plaintext %LocalAppData%\Supervertaler.Trados\settings.json ``` You can also manually back up or edit this file. After an import, the previous settings are saved as `settings.backup.json` in the same folder. # Project Settings Supervertaler for Trados automatically saves and restores your termbase configuration when you switch between Trados projects. This means each project can use its own Supervertaler database, write targets, and termbase settings without manual reconfiguration. ## How it works When you open a different Trados project (or switch to a document from another project), the plugin: 1. **Saves** the current project’s settings to a project-specific file 2. **Loads** the new project’s settings (if they exist) 3. **Reloads** the termbase with the new configuration If no project-specific settings exist yet (first time opening a project), the current global settings are used. Once you make any changes and click OK in Settings, those settings are saved for that project. Note **No action needed:** Per-project settings work automatically in the background. Just configure your termbases as usual – the plugin remembers your choices per project. ## What’s saved per project | Setting | Saved per project? | Notes | | ---------------------------------------- | ------------------ | ------------------------------------------------------------------------------------------------------------------------------ | | Supervertaler database path | Yes | Each project can use a different `.db` file | | Enabled/disabled termbases (Read toggle) | Yes | Different termbases active per project | | Write targets | Yes | Different write targets per project | | Project termbase (pink highlighting) | Yes | Different project termbase per project | | MultiTerm visibility | Yes | Different MultiTerm termbases enabled per project | | AI context termbase filters | Yes | Different termbases in AI prompts per project | | Active prompt | Yes | Each project remembers its [active prompt](/trados/ai-assistant/super-memory/active-prompt/) for Quick Add and Batch Translate | | API keys and provider settings | No | Shared across all projects | | Panel font size | No | UI preference, shared | | Term shortcut style | No | UI preference, shared | | Dialogue sizes | No | UI layout, shared | ## Storage location Per-project settings are stored as individual JSON files inside your [user data folder](/trados/data-folder/): ```plaintext C:\Users\{you}\Supervertaler\trados\projects\ ``` Each file is named with a hash and the project name (e.g., `a1b2c3d4 - MyProject.json`). The JSON file also contains the original project path for reference. Caution **Moving a Trados project** to a different folder creates a new project key. The plugin will treat it as a new project and use global defaults until you reconfigure. The old project settings file remains in the `projects` folder and can be safely deleted. ## Interaction with global settings Global settings (`settings.json`) serve as the defaults for projects that don’t have their own settings file yet. When you open a project for the first time, the global termbase configuration is used. Once you change settings and click OK, those settings are saved for that specific project. Settings that are always global (API keys, font size, shortcut preferences) are never overridden by project settings. *** ## See Also * [TermLens Settings](/trados/settings/termlens/) * [AI Settings](/trados/settings/ai-settings/) * [Backup & Restore](/trados/settings/backup/) # Prompts Prompts tell the AI how to behave. The Prompt Manager lets you browse built-in domain prompts, create your own, and mark prompts as QuickLauncher shortcuts. #### Accessing the Prompt Manager Open the plugin **Settings** dialogueue and switch to the **Prompts** tab. ![]() #### Active prompt Right-click any translation prompt in the tree and choose **Set as active prompt for this project** to designate it as the active prompt. The active prompt is shown with a pin icon and bold blue text. It is used by [memory bank Quick Add](/trados/ai-assistant/super-memory/quick-add/) when appending terminology, and is auto-selected in the [Batch Translate](/trados/batch-translate/) dropdown. The active prompt is saved [per project](/trados/settings/project-settings/). #### What’s in this section * [**Built-in Prompts**](/trados/settings/prompts/built-in-prompts/) – the prompts that ship with the plugin, organised by category * [**Prompt Variables**](/trados/settings/prompts/prompt-variables/) – placeholders you can use to make prompts context-aware (language, segment, project) * [**Writing Custom Prompts**](/trados/settings/prompts/writing-custom-prompts/) – anatomy of a good prompt, worked examples, and tips * [**QuickLauncher Shortcuts**](/trados/settings/prompts/quicklauncher-shortcuts/) – marking prompts as QuickLauncher items, keyboard shortcuts (Ctrl+Alt+1-0), and reordering * [**Organising Prompts**](/trados/settings/prompts/organising-prompts/) – folders, drag-and-drop, and the prompt library structure * [**Prompt File Format**](/trados/settings/prompts/prompt-file-format/) – the `.svprompt` format, YAML fields, and creating/editing/deleting prompts #### See Also * [AutoPrompt](/trados/generate-prompt/) * [QuickLauncher](/trados/quicklauncher/) * [AI Settings](/trados/settings/ai-settings/) * [Batch Translate](/trados/batch-translate/) * [AI Proofreader](/trados/ai-proofreader/) * [SuperMemory](/trados/ai-assistant/super-memory/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) # Built In Prompts The plugin ships with default prompts organised into three categories: | Category | Prompts | Used in | | ----------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------- | | **Translate** | Default Translation Prompt | Batch Translate mode | | **Proofread** | Default Proofreading Prompt | Batch Proofread mode | | **QuickLauncher** | Assess translation, Define, Explain (in general), Explain (within project context), Translate segment using fuzzy matches, Translate selection in context of current project | QuickLauncher menu | Note The Default Translation Prompt is a general-purpose starting point. For domain-specific projects, use **[AutoPrompt](/trados/generate-prompt/)** to automatically create a comprehensive prompt tailored to your document – or duplicate the default prompt in the Prompt Manager and customise it manually. ### About the Default Proofreading Prompt The Default Proofreading Prompt is intentionally short. Most of the structure proofreading needs – persona, the five quality categories (accuracy / completeness / terminology / grammar / number formatting), the output format, the “no full corrected translations” rule, language-specific checks for Dutch / German / French – is **already in the hardcoded base** that every Batch Proofread uses, so the prompt itself only adds what the base doesn’t have: * **Default to OK** – raise an issue only when there’s a specific, demonstrable problem in the translation, never speculative concerns. * **Citation discipline** – terminology consistency claims must cite specific source segment numbers in the **Evidence:** field, against the full bilingual document context that’s auto-included. * **Source query distinction** – source-side errors (typos, duplications, internal inconsistencies) get prefixed with “Source query:” rather than triggering target changes. * **Explicit boundaries** – the AI doesn’t re-engineer the source, propose alternative terminology without a citation, flag stylistic preferences, or flag empty target lines. These behaviours target the false-positive patterns most users encounter: the AI fabricating “term X used elsewhere” claims, second-guessing the source’s substantive claims, and treating stylistic preferences as errors. See [AI Proofreader](/trados/ai-proofreader/) for the full picture, including the Evidence field on issue cards. Note **To customise:** clone the default in the Prompt Manager and edit your copy. Defaults are read-only and get refreshed when the plugin updates – your clone keeps all your changes regardless. When the Prompt Library writes the default to disk, it includes a `default: true` flag in the YAML frontmatter; clones get `default: false` and are never touched by future updates. # Organising Prompts The Prompt Manager tree mirrors the folder structure inside your `prompt_library` directory. You can: * **Create folders** – click **New Folder** in the toolbar * **Move prompts** – drag and drop a prompt onto a folder to move it * **Clone prompts** – right-click any prompt and select **Clone** to create a copy with “(2)” appended to the name, in the same folder as the original * **Delete prompts** – right-click a prompt or folder and select **Delete** * **Browse prompts** – click any prompt to preview its content in the detail pane # Prompt File Format Prompts are stored as `.md` files (Markdown with YAML frontmatter). This is the same format used by Supervertaler Workbench, so prompts are automatically shared between both applications via the shared `prompt_library` folder. Legacy `.svprompt` files are still loaded for backward compatibility. ```yaml --- type: prompt description: Patent and IP translation with strict terminology rules category: Translate --- You are an expert {{SOURCE_LANGUAGE}} to {{TARGET_LANGUAGE}} patent translator... ``` The example file above would be saved as e.g. `My Patent Prompt.md` and would appear in the prompt tree as **My Patent Prompt**. ### Naming: filename is authoritative The **on-disk filename** (without the `.md` extension) is the display name shown in the prompt selector. Renaming `My Patent Prompt.md` to `Client X – Patent EN.md` in Windows Explorer is all you need to do to change how the prompt appears in the tree – click **Refresh** in the Prompts tab to pick up the change. Note **The YAML `name:` field is ignored on read.** It used to be the authoritative display name, but that created a confusing split: renaming the file in Explorer didn’t update the tree unless you also edited the YAML inside. Filename is now the single source of truth. Old prompts with a `name:` field in their YAML continue to load fine – the field is silently ignored, and is dropped from the file the next time the prompt is saved through the UI. No action required for existing prompts. | YAML field | Description | | --------------------- | -------------------------------------------------------------------------------------------------- | | `type` | Document type – always `prompt` for prompt files | | `description` | Optional summary shown under the prompt name in the detail pane | | `category` | `Translate`, `Proofread`, or `QuickLauncher` – controls where the prompt appears | | `quicklauncher_label` | Short label for the QuickLauncher menu (optional, falls back to the filename) | | `default` | `true` for shipped prompts (managed by the plugin) | | `sort_order` | Numeric order within folder (lower values first). Set automatically by the ▲/▼ buttons. | | `name` | Ignored on read (legacy field, kept for backward compatibility). The filename is the display name. | Note Older prompts using the `domain` key instead of `category` are still supported for backward compatibility. ### System prompt The plugin automatically prepends a system prompt to every AI call. This system prompt includes language pair information, termbase terms (based on your [AI Context settings](/trados/settings/ai-settings/)), and TM matches when enabled. The content you write in a prompt `.md` file is the **user prompt** – it is sent after the system prompt. ### Creating and editing prompts #### New prompt 1. Optionally select a target folder (e.g. `Translate` or `Proofread`) in the tree before clicking **New** – the new prompt’s **Category** will be pre-filled from the selected folder. 2. Click **New** in the Prompts tab. 3. Fill in Name, Description, Category, and Content. 4. Click **Save**. Note **Category matters for Batch Translate.** The Batch Translate dropdown filters by category: Translate mode only shows prompts whose Category is `Translate`, and Proofread mode only shows `Proofread` prompts. If you click **New** without first selecting a folder, the category defaults to `Translate` so the new prompt is immediately visible in the Batch Translate dropdown. Prompts with an empty or unrelated category will not appear in either Batch mode – move them into a `Translate` or `Proofread` folder (or edit the Category field) to make them selectable. #### Edit a prompt 1. Select a prompt in the list 2. Click **Edit** 3. Modify as needed and click **Save** #### Inserting variables While editing prompt content, press **Ctrl+,** to open the variable picker menu. This lists all available variables with a short description. Select a variable to insert it at the cursor position. If text is selected in the editor, it is replaced by the inserted variable. Note **Ctrl+,** mirrors the variable insertion shortcut used in the Trados Studio editor. #### Delete a prompt 1. Select a custom prompt 2. Click **Delete** and confirm Built-in prompts cannot be deleted. Click **Restore** to recreate any built-in prompts you have removed. # Prompt Variables Variables are placeholders in your prompt text that are automatically filled in at runtime. Use them to make prompts context-aware without rewriting them for every project or language pair. ### Language variables – all contexts These work in Batch Translate, Batch Proofread, and QuickLauncher prompts: | Variable | Replaced with | Example | | --------------------- | -------------------------------------------------- | ------------------------- | | `{{SOURCE_LANGUAGE}}` | Full name of the source language, including locale | `Dutch (Belgium)` | | `{{TARGET_LANGUAGE}}` | Full name of the target language, including locale | `English (United States)` | ### Segment variables – QuickLauncher only These are only available in QuickLauncher prompts, because they refer to the specific segment active at the moment you trigger the menu: | Variable | Replaced with | Example | | -------------------- | ---------------------------------------------------------------------------------------------- | ----------------------------------------------------------- | | `{{SOURCE_SEGMENT}}` | Full source text of the **active segment** | `De uitvinding heeft betrekking op een nieuwe werkwijze...` | | `{{TARGET_SEGMENT}}` | Full target text of the **active segment** (your translation so far – may be empty or partial) | `The invention relates to a novel method...` | | `{{SELECTION}}` | Text currently **selected** in the editor (source side preferred; falls back to target side) | `werkwijze` | Note **Segment vs selection:** `{{SOURCE_SEGMENT}}` and `{{TARGET_SEGMENT}}` always give you the **entire active segment**. `{{SELECTION}}` gives you only the **highlighted portion** – useful for looking up or explaining a specific word or phrase within the segment. If nothing is selected, `{{SELECTION}}` is replaced with an empty string. ### Project variables – QuickLauncher only | Variable | Replaced with | | -------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `{{PROJECT_NAME}}` | Trados project name (e.g. `Patent_NL_EN_2026`) | | `{{DOCUMENT_NAME}}` | Active file name (e.g. `source_document.docx`) | | `{{SURROUNDING_SEGMENTS}}` | N source segments before and after the active segment, with actual Trados segment numbers and the active segment marked `← ACTIVE`. N is set in **Settings → AI Settings → Surrounding segments** (default: 5). | | `{{PROJECT}}` | All source segments in the document, numbered with their actual Trados segment numbers. In multi-file projects a `=== File N ===` header separates each file (Trados restarts segment numbering per file). | | `{{TM_MATCHES}}` | Translation memory fuzzy matches (≥70%) for the active segment, showing match percentage, TM name, source text, and target text. If no matches meet the threshold, replaced with “(no fuzzy matches above 70%)”. | Caution `{{PROJECT}}` sends the entire document to the AI and uses significantly more tokens than other variables. For a 10,000-word document, this costs roughly 4–5 cents per call with a Sonnet-class model. Reserve it for prompts where full document context genuinely matters. To keep the chat history readable, the chat bubble shows a compact summary (e.g. `[source document – 47 segments]`) instead of the full source text. The complete document is still sent to the AI. ### Scope: which prompts see which variables The variables above are **only substituted in QuickLauncher prompts**. In Batch Translate and Batch Proofread custom prompts, only the two **language variables** (`{{SOURCE_LANGUAGE}}` and `{{TARGET_LANGUAGE}}`) get filled in – any other variable left in the prompt body will be replaced with an empty string. This isn’t a limitation in practice. The batch flows assemble a richer system prompt automatically, with no variables required: * **Batch Translate** automatically includes source-only document context (when **Include document context** is on in [AI Settings](/trados/settings/ai-settings/)), termbase entries, and language-specific checks. * **Batch Proofread** automatically includes the **full bilingual document context** (source + target, untruncated), termbase entries, and language-specific checks. So the right place for your custom prompt content in batch mode is to **complement** what’s already there – domain-specific guidance, citation discipline, project-specific terminology preferences – rather than try to re-inject document context with `{{PROJECT}}`. If you want to see exactly what gets sent to the AI for a batch run – fully assembled, including the auto-injected context – click the **👁 Preview prompt** link next to the Translate / Proofread button on the Batch Operations tab. See [Batch Operations](/trados/batch-operations/) for details. # Quicklauncher Shortcuts ### Marking a prompt as a QuickLauncher shortcut To make a custom prompt appear in the QuickLauncher right-click menu (`Alt+Q`), set `category: QuickLauncher` in the YAML frontmatter: ```yaml --- name: Explain selected term description: Explains the selected term in translation context category: QuickLauncher quicklauncher_label: Explain term --- Your prompt content here... ``` | Field | Description | | ------------------------- | ------------------------------------------------------------------------ | | `category: QuickLauncher` | Marks this prompt as a QuickLauncher item | | `quicklauncher_label` | Optional short label shown in the menu – falls back to `name` if omitted | You can also organise QuickLauncher prompts by placing them in a folder called `QuickLauncher` inside your `prompt_library` folder. Any prompt in that folder is automatically treated as a QuickLauncher prompt. Note QuickLauncher prompts are shared with Supervertaler Workbench via the shared prompt library folder. ### Keyboard shortcuts for QuickLauncher prompts You can assign keyboard shortcuts (Ctrl+Alt+1 through Ctrl+Alt+0) to individual QuickLauncher prompts for instant access without opening the Alt+Q menu. 1. Open **Settings → Prompts** 2. Select a QuickLauncher prompt in the tree 3. In the detail pane on the right, use the **Shortcut** dropdown to assign a slot 4. Click **OK** to save Each shortcut can only be assigned to one prompt. If you assign a shortcut that is already in use, it is automatically cleared from the other prompt. Assigned shortcuts are shown next to prompt names in the Alt+Q menu and in the Trados keyboard shortcuts settings (File → Options → Keyboard Shortcuts → Supervertaler for Trados). ### Reordering prompts Use the **▲** and **▼** buttons in the toolbar to change the order of prompts within a folder. This is especially useful for QuickLauncher prompts, as the order in the tree determines the order in the Alt+Q menu. The order is saved in each prompt’s YAML frontmatter as a `sort_order` field. # Writing Custom Prompts ### Anatomy of a prompt A good prompt has three parts: 1. **Role** – tells the AI who it is 2. **Task** – tells the AI what to do 3. **Constraints** – tells the AI what to avoid or preserve Here is an annotated example for a Batch Translate prompt: ```plaintext You are an expert {{SOURCE_LANGUAGE}} to {{TARGET_LANGUAGE}} patent translator. ← Role + language variables Translate the source segment provided. Return only the translated text – ← Task no commentary, no explanations, no repetition of the source. Preserve all tag placeholders exactly as they appear (e.g. , ). ← Constraints Preserve numbers, units, and chemical formulas without conversion. Use formal, technical register throughout. ``` ### Example QuickLauncher prompt – explain a selected term This prompt uses `{{SELECTION}}` to ask the AI to explain a selected term in context: ```plaintext The user is translating a patent from {{SOURCE_LANGUAGE}} to {{TARGET_LANGUAGE}}. The selected term is: {{SELECTION}} Please explain what this term means in the context of patent translation, suggest the standard {{TARGET_LANGUAGE}} equivalent, and note any regional or register variations the translator should be aware of. ``` When the translator selects “werkwijze” in the source segment and triggers this prompt via QuickLauncher, the AI receives: ```plaintext The user is translating a patent from Dutch (Belgium) to English (United States). The selected term is: werkwijze Please explain what this term means... ``` ### Example QuickLauncher prompt – assess the current translation This prompt uses `{{SOURCE_SEGMENT}}` and `{{TARGET_SEGMENT}}` to ask the AI to review the translation of the active segment: ```plaintext Source ({{SOURCE_LANGUAGE}}): {{SOURCE_SEGMENT}} My translation ({{TARGET_LANGUAGE}}): {{TARGET_SEGMENT}} Assess how I translated the current segment. Point out any inaccuracies, awkward phrasing, or terminology issues, and suggest improvements. ``` ### Example QuickLauncher prompt – translate a selected term in context ```plaintext Source segment ({{SOURCE_LANGUAGE}}): {{SOURCE_SEGMENT}} Current translation ({{TARGET_LANGUAGE}}): {{TARGET_SEGMENT}} The translator has selected this word or phrase: {{SELECTION}} Suggest the best {{TARGET_LANGUAGE}} translation for "{{SELECTION}}" given the full segment context above. Give a short explanation of your reasoning. ``` ### Example QuickLauncher prompt – translate a term using surrounding passage Uses `{{SURROUNDING_SEGMENTS}}` for a wider context window than just the active segment: ```plaintext I am translating a {{SOURCE_LANGUAGE}} patent into {{TARGET_LANGUAGE}}. The selected term is: {{SELECTION}} Here is the passage surrounding the active segment: {{SURROUNDING_SEGMENTS}} Suggest the best {{TARGET_LANGUAGE}} translation for "{{SELECTION}}" given the surrounding context. Briefly explain your reasoning. ``` ### Example QuickLauncher prompt – full-document term consistency check Uses `{{PROJECT}}` to give the AI the entire source document. Useful for checking whether a key term is used consistently, or for understanding a term’s meaning across all its occurrences. Reserve this for important queries – see the token cost note in [Prompt Variables](/trados/settings/prompts/prompt-variables/). ```plaintext I am translating a {{SOURCE_LANGUAGE}} patent ({{DOCUMENT_NAME}}) into {{TARGET_LANGUAGE}}. Project: {{PROJECT_NAME}} Here is the complete source text: {{PROJECT}} The selected term is: {{SELECTION}} What is the most accurate and consistent {{TARGET_LANGUAGE}} translation for "{{SELECTION}}" throughout this document? Note any variation in meaning between occurrences and recommend which translation to use where. ``` ### Example QuickLauncher prompt – check a segment against the full document After sending `{{PROJECT}}`, the AI knows the segment numbers shown in Trados, so you can ask about specific segments by number in follow-up messages – or ask in the prompt itself: ```plaintext I am translating a {{SOURCE_LANGUAGE}} patent into {{TARGET_LANGUAGE}}. Here is the source document: {{PROJECT}} I am currently working on segment {{SOURCE_SEGMENT}} (shown as [{{SOURCE_SEGMENT}}] above). My translation is: {{TARGET_SEGMENT}} Does this translation accurately reflect the source and maintain consistency with the terminology used elsewhere in the document? Point out any issues. ``` ### Tips for effective prompts * **Be explicit about output format.** If you only want the translation, say “Return only the translated text.” If you want an explanation, describe the expected structure. * **Use language variables.** Hardcoding “Dutch to English” breaks the prompt when you switch projects. Always use `{{SOURCE_LANGUAGE}}` and `{{TARGET_LANGUAGE}}`. * **Keep QuickLauncher prompts focused.** A narrow, specific task works better than a broad one – except when you deliberately need the full document context via `{{PROJECT}}`. * **Use `{{SURROUNDING_SEGMENTS}}` instead of `{{SOURCE_SEGMENT}}` when context matters.** The surrounding passage often gives the AI enough context for a better answer at a fraction of the cost of `{{PROJECT}}`. * **Use `{{PROJECT}}` sparingly.** It is best suited for high-stakes queries on short-to-medium documents – terminology consistency checks, key term decisions, or reviewing a handful of specific segments. Avoid it in prompts you run on every segment. * **Segment numbers in `{{PROJECT}}` match the Trados editor.** After sending `{{PROJECT}}`, you can ask the AI about “segment 4” or “segment 12” and it will know exactly which segment you mean – the same number shown in the Trados grid. * **Use `{{TM_MATCHES}}` to leverage existing translations.** When a segment has a high fuzzy match, the AI can use it as a starting point – especially useful for repetitive or formulaic content like patents and legal texts. * **Batch Translate prompts receive one segment at a time.** You do not need to handle lists of segments or loop logic. * **Proofread prompts receive multiple segment pairs.** The built-in proofreading prompt shows the expected input/output format – follow that structure if you write a custom one. # Termlens Configure how TermLens loads and displays terminology in Trados Studio. ## Accessing TermLens settings Click the **gear icon** in the TermLens panel, or open the plugin **Settings** dialogue and switch to the **TermLens** tab. ## Database path The path to your Supervertaler termbase `.db` file. Click **Browse** to select a database, or **Create New** to start with an empty one. Note **Auto-detect:** If Supervertaler Workbench is installed on the same machine, the plugin can automatically detect its default database location. Click **Auto-detect** to find and use it. ## Termbase toggles Each Supervertaler termbase in the database has three toggles. See [Termbase Management](/trados/termbase-management/) for full details. | Toggle | Purpose | | ----------- | ------------------------------------------------------------------------------------- | | **Read** | Load terms for matching –only termbases with Read enabled appear in TermLens | | **Write** | Receive new terms added via the [quick-add shortcuts](/trados/termlens/adding-terms/) | | **Project** | Mark as the project termbase (shown in pink, prioritised) | ### Confirm dialog for non-matching termbases When you tick **Write** or **Project** on a termbase whose declared language pair does not match the active project (for example, ticking an EN→NL termbase as Write while you have a DE→FR project open), a confirmation dialog appears: > *“\” is a EN → NL termbase, but the active project’s source language is German. Setting it as a Write termbase means new terms added during this project will be written into a termbase whose language pair doesn’t match.* > > *This is occasionally intentional (multilingual or global termbases, bootstrapping a new direction) – tick “Yes” to continue. The plugin will remember this choice for this termbase and won’t ask again until you untick the box.* Click **Yes** to keep the tick, or **No** to revert it. The plugin remembers each “Yes” answer per termbase, so once you have explicitly confirmed a non-matching termbase you are not asked again on subsequent ticks. **Unticking the box clears that confirmation** – a future re-tick will re-ask. This is intentional: an untick is taken as a clear signal that you are reconsidering, so the next tick deserves a fresh look. The header **tick-all** on the Write column also respects this guard – non-matching termbases that haven’t been individually confirmed are skipped during a bulk tick, so a quick “tick everything” can’t accidentally enable unrelated termbases for write. The **Read** column is intentionally exempt from the dialog – there is no harm in *reading* a non-matching termbase (its terms simply won’t match anything in your segments), only in writing into it. Note **No project loaded.** If you open the Settings dialog without an active project (for example by clicking the gear from the QuickLauncher header), the plugin has no source language to compare against, so the confirmation is suppressed and ticks behave normally. ## MultiTerm termbases If your Trados project has MultiTerm termbases (`.sdltb` files) attached, they appear at the bottom of the termbase list with a **\[MultiTerm]** label and a light green row background. The **Read** toggle controls visibility in TermLens; **Write** and **Project** are always disabled because MultiTerm termbases are read-only. To add or remove MultiTerm termbases, use Trados Studio’s **Project Settings > Language Pairs > Termbases**. See [MultiTerm Support](/trados/multiterm-support/) for full details. ## Auto-load on startup When enabled, the plugin automatically loads the termbase database when Trados Studio opens. This means terms are available immediately when you start translating, without needing to open the settings first. If disabled, the termbase loads the first time you open the TermLens settings or click the TermLens panel. ## Case-sensitive matching By default, TermLens matches terms regardless of letter case – “polymer”, “Polymer”, and “POLYMER” all match the same term entry. Enable **“Enable case-sensitive matching globally”** to require exact case matching across all termbases. You can also control case sensitivity per termbase using the **CS** checkbox in the termbase grid. When the CS checkbox is ticked for a termbase, that termbase always matches case-sensitively; when unticked, it matches case-insensitively. Note **Tip:** The CS checkbox is useful when you have one termbase with abbreviations that must match exactly (e.g., “GC” should not match “gc”) while other termbases should remain case-insensitive. ## Adapt term capitalisation **Adapt term capitalisation to the segment** (on by default) makes TermLens display and insert target terms with the capitalisation of the source occurrence in the segment rather than the capitalisation stored in the termbase: * A term stored as “More preferably” shows and inserts as **“more preferably”** when the segment contains it lower-case mid-sentence * A term stored lower-case is **capitalised** when the occurrence starts the sentence * An **ALL-CAPS** occurrence (e.g. in a heading) upper-cases the whole inserted term The adaptation applies everywhere a term is displayed or inserted: the TermLens chips, Alt+digit shortcuts, the TermLens popup and TermPicker. Note The rules are deliberately conservative. Acronyms and mixed-case terms (MRI, pH, MultiTerm) are never altered, abbreviation matches keep their stored casing, and suffix-tolerant Korean/Japanese matches are left untouched. Untick the option to always show and insert terms exactly as stored. ## Panel font size Adjust the font size used in the TermLens display panel. Valid range: **7 pt** to **16 pt**. Increase the font size if TermLens text is hard to read; decrease it to fit more terms on screen. ## Term shortcuts Choose how Alt+digit shortcuts work when a segment has more than 9 matched terms: * **Sequential** (default) – type the term number digit by digit. Alt+45 inserts term 45. Badges show clean sequential numbers (10, 11, 12, …). There is a brief delay after each digit while the system waits for a possible next digit. * **Repeated digit** – press the same digit key multiple times. Alt+55 inserts term 14 (the 5th term in the second tier). Badges show repeated digits (11, 22, 333, …). No delay, but the badges are less intuitive. Both modes behave identically when a segment has 9 or fewer matches – pressing Alt+N inserts immediately with no delay. ## Shortcut delay Controls how long the system waits for the next digit in **Sequential** mode (in milliseconds). Default: **1100 ms**. Valid range: **300 ms** to **3000 ms**. Increase the delay if you need more time between keystrokes. Decrease it if you find the pause too long when inserting single-digit terms in segments with 10+ matches. This setting has no effect in Repeated digit mode. See [Keyboard Shortcuts](/trados/keyboard-shortcuts/) for the full reference. *** ## See Also * [Termbase Management](/trados/termbase-management/) * [MultiTerm Support](/trados/multiterm-support/) * [AI Settings](/trados/settings/ai-settings/) * [TermLens overview](/trados/termlens/) # Usage Statistics Supervertaler for Trados sends one anonymous, lightweight ping to the developer at startup so he can see how many people are using the plugin and what environments they are running it on. The dialogue below appears once after install or update, and the feature can be switched off at any time.  #### How it works * **Default-on, opt-out** – on first launch after install or update, an informational dialogue (shown above) tells you exactly what is collected and gives you a one-click **Turn it off** button. You don’t have to do anything to keep it enabled – the dialogue’s default action (the bold **Keep it on** button, Enter, Esc, or the X-close) all keep it enabled. Your choice is remembered, and the dialogue isn’t shown again. * **Minimal data** – a single lightweight ping is sent once per session on plugin startup. The only data included is: * A random anonymous ID (a UUID generated locally on your machine – not tied to any account, machine, or identity) * Plugin version (e.g. 4.19.108) * OS version (e.g. Windows 11) * Trados Studio version * System locale (e.g. en-GB) * **Country detection** – the hosting provider (Cloudflare) determines your country from the network connection. No IP addresses are stored. * **Silent failure** – if the ping fails (no internet, firewall, etc.), nothing happens. No retries, no queuing, no error messages. * **First-party only** – data is sent to a Supervertaler-operated Cloudflare Worker endpoint. No third-party trackers, no Google Analytics, no advertising platforms. #### What is NOT collected * No translation content * No termbase data * No file names or project names * No personal information (name, email, etc.) * No information about which features you use or how often * No API keys or credentials #### Changing your preference You can change your choice at any time: 1. Open **Settings** (click the gear icon in the Supervertaler Assistant panel) 2. In the **TermLens** tab, scroll to the **Privacy** section 3. Check or uncheck **“Share anonymous usage statistics (no personal data)”** 4. Click **OK** The change takes effect on the next Trados Studio session. #### Why this exists As a solo developer, usage statistics provide invaluable insight into: * How many people are actually using the plugin * Which Trados Studio versions to prioritise for testing and compatibility * Which OS versions and locales are most common * Whether users run Trados on a Mac (via Parallels) or natively on Windows This information directly informs development priorities and compatibility testing. #### Transparency The full source code for both the plugin-side statistics and the server-side endpoint is publicly available: * **Plugin code**: [`Core/UsageStatistics.cs`](https://github.com/Supervertaler/Supervertaler-for-Trados/blob/main/src/Supervertaler.Trados/Core/UsageStatistics.cs) on GitHub * **Server code**: The Cloudflare Worker that receives the pings is also open source You can verify exactly what data is sent by inspecting the code yourself. # Trados Studio 2026 & .ttb Termbases Trados Studio 2026 introduces a new termbase format –the SQLite-based **`.ttb`** file –and drops the legacy MultiTerm engine that earlier versions relied on. Supervertaler for Trados supports Studio 2026 through a dedicated build that reads `.ttb` termbases directly. ## Two builds, one product Supervertaler for Trados ships as two separate plugin builds from the same codebase: | Build | For | Termbase format | | ------------------------------------------ | ------------------ | ------------------ | | **Supervertaler for Trados** | Trados Studio 2024 | MultiTerm `.sdltb` | | **Supervertaler for Trados (Studio 2026)** | Trados Studio 2026 | `.ttb` | Install the build that matches your Studio version. The 2024 build will not load in Studio 2026, and vice versa, because the two Studio releases use different plugin frameworks and termbase engines. ## Version numbering Each build’s **major version tracks the Trados Studio major it targets**, so the two builds always carry distinct, non-colliding version numbers that share the same tail: | Build | Example version | | ------------------ | --------------- | | Trados Studio 2024 | `18.20.86` | | Trados Studio 2026 | `19.20.86` | So you can tell at a glance which Studio a build is for: a `18.x` plugin is for Studio 2024, a `19.x` plugin is for Studio 2026. (Releases up to and including `4.20.85` used a single shared number for both builds.) ## TermLens and `.ttb` termbases In Studio 2026, TermLens reads the new `.ttb` termbases attached to your project automatically –there is nothing to configure. Terms appear as **green chips** in the TermLens panel, exactly as MultiTerm terms do in the 2024 build, and behave the same way (click to insert, or **Alt+1**–**Alt+9** to insert by number). `.ttb` termbases are **read-only** in TermLens, just like MultiTerm termbases. To add or edit terms, use Studio 2026’s built-in Termbases view. Everything else – Supervertaler, Batch Translate, SuperSearch, TermPicker, AI terminology injection – works identically across both builds. ## What about my existing `.sdltb` termbases? Studio 2026 cannot open `.sdltb` files directly; the MultiTerm engine that read them is no longer part of the base install. Instead, **RWS provides conversion to the new `.ttb` format**: * The **Termbases view** in Studio 2026 has a wizard to migrate `.sdltb` → `.ttb`. * Opening a project package that contains an `.sdltb` termbase converts it to `.ttb` on the fly. * When a 2026 user sends a package back to someone still on Studio 2024, Studio can create a “compatible” package that the older version converts back to `.sdltb` automatically. Once your termbase is in `.ttb` form, TermLens picks it up automatically. There is no separate step in Supervertaler –convert in Studio, and the terms appear. Note If you work in both Studio 2024 and Studio 2026, keep using the MultiTerm `.sdltb` build for your 2024 projects. The 2026 build is only for Studio 2026 and its `.ttb` termbases. See [MultiTerm Support](/trados/multiterm-support/) for the 2024 workflow. ## Technical notes * `.ttb` is a SQLite 3 database with full-text search. TermLens opens it in read-only mode, so Studio can keep using it at the same time with no risk of corruption. * Studio 2026 is a 64-bit application. The 2026 build is 64-bit to match; the 2024 build remains 32-bit/AnyCPU for the MultiTerm JET driver it depends on. ## See also * [MultiTerm Support](/trados/multiterm-support/) –the equivalent `.sdltb` workflow in Studio 2024 * [TermLens](/trados/termlens/) * [Installation (Trados)](/trados/installation/) * [Troubleshooting](/trados/troubleshooting/) # SuperBench > Translate the same text with three models under identical settings, and have a judge say which to use for this project. Which model should you use for this project? SuperBench answers that with evidence from the document in front of you rather than from a benchmark of somebody else’s text. Available from v18.20.187. ## What it does 1. You pick **three models**, one per slot – any provider each. The defaults are Claude Opus 5, GPT-5.6 Sol and Gemini 3.1 Pro. 2. You pick a **judge** – Claude Fable 5.1 by default – and how many segments of the open document to use. Twenty is a good start. 3. Each model translates those segments through the batch pipeline itself. The selected prompt, the termbase terms that occur in the text, the document context, SuperMemory and the batch size are exactly what a real Batch Translate would send; only the model changes. **Nothing is written to the document.** 4. The judge reads the source and the three translations **blind** – labelled A, B and C in a shuffled order, so no brand name colours the verdict – together with the same approved terms and instructions, and writes a short report: a ranked verdict, the significant errors per candidate with segment numbers, how each handled terminology and tags, and a recommendation for this project. 5. The translations appear side by side above the report, with each model’s cost and time. The whole thing is saved as Markdown in `trados\reports` inside your data folder, and **Save report…** writes a copy wherever you like.  A finished run on the inline-formatting test document: Claude Opus 5, GPT-5.6 Sol and Gemini 3.1 Pro side by side, judged by Claude Fable 5.1. The legend above the report maps the judge's letters to the models. ## Running it 1. Open the document and set up the Batch Operations tab as you would for a real run: the prompt, and the termbases ticked for AI on the Termbases tab. 2. Click **⚖ SuperBench…** next to **Preview prompt**. 3. Choose the three models, the judge and the number of segments. The line under the segment count shows the estimated cost of the three runs and the judge’s call. 4. Click **Run SuperBench**. The status line follows the runs; **Cancel** stops after the current call. ## Reading the report The legend at the top of the report maps the judge’s letters to the models. The judge’s own text uses only the letters. * **Verdict** – the best candidate for this text, and a ranked list. * **Errors** – the significant errors per candidate, each with the segment number and the words in question. “None found” means the judge looked and found nothing. * **Terminology** and **Tags** – whether the approved terms were used and the inline tags kept. * **Recommendation** – which to use for this project, and whether the difference is worth paying for. A judge is a model too. Treat the report as a well-informed second opinion: read the segments it names before deciding, and run it again on a different stretch of the document if the verdict is close. ## Cost Three batches of twenty segments plus one judge call costs a few cents with the default models; the estimate is shown before you run, and the actual cost per model is in the report. Every call is logged like any other AI call, so it appears in the Reports tab and the usage ledger. # SuperSearch SuperSearch is a cross-file search and replace tool that lets you find text across **all SDLXLIFF files** in your Trados project – not just the file you currently have open – and, optionally, across the project’s **translation memories** as well. It lives in its own dockable panel, so you can keep it visible while you translate. Matching text is highlighted in yellow in the results grid, making it easy to spot exactly where the search term appears in each segment. [SuperSearch in action – cross-file search across a Trados project](https://www.youtube.com/embed/549Ulc92FiU) ## Opening the Panel There are three ways to open SuperSearch: | Method | Description | | --------------- | -------------------------------------------------------------------------- | | **View menu** | Go to **View > SuperSearch** | | **Right-click** | Right-click in the editor and choose **SuperSearch** from the context menu | | **Keyboard** | Press **Alt+S** | The panel docks at the bottom of the editor by default, but you can drag it anywhere – left, right, floating, or even to a second monitor. Trados remembers the position between sessions. Note **Prefer fewer panels?** You can host SuperSearch as a tab inside the Supervertaler Assistant panel instead of its own dockable panel. Go to **Settings > General > Panels** and tick **Show SuperSearch as a tab in the Supervertaler Assistant panel**, then restart Trados Studio. This requires a Supervertaler licence; without one, SuperSearch stays in its own panel. Note **Quick search from the editor:** Select a word or phrase in the source or target segment, then press **Alt+S** (or right-click > **SuperSearch**). The selected text is automatically entered in the search box and the search runs immediately. ![]() ## Searching Type your search query in the text box and press **Enter** (or click **Search**). ### Search Modes The **mode** dropdown in the search bar controls where SuperSearch looks: | Mode | Searches | | ----------------- | ------------------------------------------------------------------------------------------------------ | | **Everything** | Project files, translation memories *and* termbases, merged into one result list | | **Project files** | The project’s SDLXLIFF files | | **TMs** | Only the project’s translation memories – a concordance search, like Studio’s built-in Concordance | | **Termbases** | Only your terminology – Supervertaler, MultiTerm (`.sdltb`) and Trados `.ttb` termbases *(v18.20.155)* | The mode is remembered across sessions until you change it. Translation-memory results are found via the project’s attached file-based TMs (`.sdltm`) – read from the project settings and the project’s `Tm` folder. Server-based (GroupShare) TMs are not searched. TM hits obey the same **Aa**, **.\***, and **Word** options as file results, and the **Scope** dropdown maps to source-side / target-side concordance. The TM list is re-checked every time you search, so a TM you attach to the project mid-session is picked up without reopening the project. SuperSearch searches every attached TM regardless of its **Enabled** / **Concordance** state in the project’s TM settings – use the **TMs** button (see below) to narrow the list. ### Searching your termbases *(from v18.20.155)* *“Where does this phrase appear?”* and *“what have I called this term?”* are the same question at different granularities, so SuperSearch answers both. Terminology is searched alongside files and TMs in **Everything**, or on its own in **Termbases**. All three kinds of termbase are included, with nothing extra to configure: * **Supervertaler termbases** – every termbase in your shared database * **MultiTerm** – the `.sdltb` termbases attached to the Trados project * **Trados `.ttb`** – the Studio 2026 termbase format Terminology is matched **in your project’s direction**: a termbase declared the other way round (an EN→NL termbase in an NL→EN project) is oriented before matching, so the **Src** box always means *the language you translate from* – not “whichever column that particular termbase calls source”. This is the same treatment TermLens gives your terminology. Searches read the terminology TermLens already holds in memory, so a termbase search is effectively instant. (Immediately after opening a project TermLens may still be loading, in which case the first search falls back to reading the database and takes longer.) Only the termbases you have **switched on** are searched – Supervertaler termbases with their **Read** tick set, and MultiTerm/`.ttb` termbases enabled in Trados Project Settings. The Read column is your statement of which terminology applies to the job in hand, so SuperSearch honours it rather than searching everything you own. Termbase hits show the **termbase name in green** and its **kind** in the Status column (`Supervertaler`, `MultiTerm` or `TTB`). #### Editing a term from the results *(from v18.20.187)* Right-click a Supervertaler termbase hit and choose **Edit term…** to open it in the **Edit term entry** dialog – the same one TermLens opens from a chip – with the termbase already resolved. Save, and the results refresh to show the corrected term; TermLens picks the change up immediately. Right-clicking selects the row under the pointer, so you need not click it first. Two kinds of hit cannot be edited here, and the menu says so if you try: **MultiTerm** and `.ttb` entries, which are read-only throughout Supervertaler (edit them in MultiTerm), and hits from a termbase that SuperSearch read straight from the database because TermLens had not finished loading it – open that termbase in the Termbase Editor instead, or search again once TermLens has loaded. Note TM and termbase results can be read and copied (via the preview pane) but cannot be navigated to or replaced – they are reference material, not document segments. The Replace bar is therefore disabled in **TMs** and **Termbases** mode. To change a term, right-click its row and choose **Edit term…** (see below). ### Searching the web *(from v18.20.181)* The one place a translator still had to leave Studio was the web. SuperSearch now covers that too: select a term, press **Alt+W**, and every reference site you have switched on opens with the query and your project’s language pair already filled in. There is nothing to type and no language dropdown to set. You can also right-click in the editor and choose **Search the web**, or use the **🌐** button in the SuperSearch bar. Note **Alt+W is a second shortcut, not a replacement.** **Alt+S** still searches your files, TMs and termbases into the results grid exactly as before. **Alt+W** is the web half. Both can be rebound in Studio’s keyboard shortcut settings. #### Choosing which sites to search Click **Web (n)** in the SuperSearch bar – it sits beside **Files**, **TMs** and **TBs**, and works the same way. Forty-one sites ship with the plugin, of which five are on out of the box: **Beijerterm**, **IATE**, **Linguee**, **ProZ.com** and **Reverso**. The other thirty-six cover bilingual dictionaries (Glosbe, WordReference, bab.la), EU and legal terminology (EUR-Lex, EuroTermBank, Juremy, GEMET), encyclopaedic sources (Wikipedia, Wiktionary, Wikidata), English monolingual and writing references (Collins, Merriam-Webster, Oxford Collocations, SkELL, Etymonline), Dutch resources (Woordenlijst, Synoniemen.net, de Financiële Begrippenlijst), medical databases (EMA, EMC) and general search (Google, Google Patents, GitHub Code). Only five are enabled initially on purpose – forty tabs opening on your first search would be a poor introduction. Tick whichever you actually use. **Adding your own** takes a name and a URL template: `{query}` for the search term, plus `{sl}` and `{tl}` for the language codes where the site needs them. #### Where the results open A checkbox in that same dialog decides. Neither option is a fallback for the other, and both are worth having: * **In a Supervertaler window** – one window with a tab per site, reused for every search so tabs refresh in place rather than leaving a trail of windows behind. Tabs load only when you click them. * **In your own browser** – one new window containing all the tabs, which you close when you are done. Your browser brings your ad blocker and your signed-in sessions with it. #### Terms taken from the target side A term selected in the **target** is looked up **in the target language**. Searching a Dutch word in an EN→NL project queries the sites as nl→en, not en→nl – the latter is how you get a screen of nothing and conclude the site is broken. #### Sites that ask you to prove you are human Some sites, ProZ.com in particular, block embedded browsers. When that happens the tab shows a banner offering to hand the page to your own browser, where you are usually signed in and pass instantly. It is an offer rather than an automatic jump, so nothing pulls you out of the editor mid-segment – and if the check clears by itself, the page is still there underneath. ### Search Options | Option | Description | | ------------------ | ------------------------------------------------------------------------------------------------- | | **Scope** dropdown | Choose *Source & Target* (default), *Source only*, or *Target only* | | **Aa** checkbox | Case-sensitive search – when unchecked, “Hello” matches “hello”, “HELLO”, etc. | | **.\*** checkbox | Treat the query as a regular expression (see [Regex tips](/trados/supersearch/#regex-tips) below) | | **Word** checkbox | Match whole words only – “cat” won’t match “category” or “scatter”. Ignored when **.\*** is on | SuperSearch displays all matching segments in the results grid. The status bar shows the number of results, what was searched (files and/or TMs), and how long the search took. ### Results Grid Each row shows one matching segment, one TM entry, or one termbase entry: | Column | Description | | ------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Found in** | The project-file name, the translation-memory name (blue), or the termbase name (green). Hover for the full path | | **#** | Segment number within the file; for TM results, the concordance match score; empty for termbase entries | | **Source** | Source text – matching text is highlighted in yellow | | **Target** | Target text – matching text is highlighted in yellow | | **Status** | Confirmation status (Not Translated, Draft, Translated, etc.), “TM” for TM results, or the termbase kind (`Supervertaler`, `MultiTerm`, `TTB`) for terminology | ### Preview Pane Below the results grid is a preview pane showing the **full source and target text** of the selected result, side by side, with the match highlighted in yellow. This is handy when a segment is too long to read in its grid row. Click any result row to update the preview, and drag the splitter bar between the grid and the preview pane to resize it. The text in both preview boxes is **selectable**: drag to select, press **Ctrl+C** to copy, or right-click for a menu with **Copy**, **Select All**, **Copy source**, and **Copy target**. This makes it easy to reuse a previous translation verbatim – select the target phrase and paste it straight into your active segment. ## File, TM and Termbase Selection Four buttons in the search bar let you narrow what SuperSearch looks at – **Files** for the project’s SDLXLIFF files, **TMs** for the project’s translation memories, **TBs** for your termbases *(v18.20.155)*, and **Web** for reference sites *(v18.20.181)*. Each button shows how many items are included: * **Files (16)** – all 16 files in the project are included * **Files (12/16)** – 12 out of 16 files are included (4 excluded) * **TMs (3)** – all 3 project TMs are included * **TMs (1/3)** – 1 of 3 TMs is included (2 excluded) * **TBs (2)** – both available termbases are included * **Web (5/41)** – 5 of the 41 available web resources are switched on Files, TMs and termbases are all discovered when the project opens, so the counts are filled in before your first search. Web resources are your own standing choice rather than a property of the project, so that count is the same in every job until you change it. Click any button to open its selection dialog: 1. A list shows all the files (or TMs, or termbases) found, with checkboxes 2. **Check** the items you want to include in the search 3. **Uncheck** the items you want to exclude 4. Use **Select All** or **Select None** to quickly toggle everything 5. Click **OK** to apply The **Files** filter applies in **Project files** and **Everything** modes; the **TMs** filter applies in **TMs** and **Everything** modes; the **TBs** filter applies in **Termbases** and **Everything** modes. Termbases are listed as *name (kind)* – for example `BEIJER (Supervertaler)` – so two termbases that share a name remain distinguishable. Note These selections persist for the current session. When you switch to a different project, all files, TMs and termbases are included again by default. ## Navigating to a Segment **Double-click** a row (or select it and press **Enter**) to jump to that segment in the editor. * If the segment is in the **currently active file**, Trados navigates to it directly. * If the segment is in a **different file**, SuperSearch attempts to switch to that file and navigate to the segment. If the file is not loaded in the editor, you may need to open it first. * **TM results** can’t be navigated to – they aren’t document segments. Double-clicking a TM row just reminds you to use the preview pane to copy the text. ## Find & Replace Tick the **Replace** checkbox to reveal the replace bar. Replace always operates on **target text only** – source text is never modified. | Action | Description | | --------------- | -------------------------------------------------------------------------------------------------------------------------------------- | | **Replace** | Replaces the match in the currently selected result. The segment must be in the active file – double-click it first to navigate there. | | **Replace All** | Replaces all target matches across all files. A confirmation dialog shows how many segments in how many files will be affected. | ### How Replace All works * For the **active file**: changes go through the Trados API, so they appear immediately and are tracked in Trados’s undo history. * For **other files**: the SDLXLIFF XML is modified directly on disk. You need to reopen those files to see the changes. Caution **Replace All cannot be undone** for files modified on disk. Always review the search results carefully before replacing. Consider saving your project first. Note Replace respects the same **Aa** (case sensitivity) and **.\*** (regex) settings as search. When using regex, you can use capture groups in the replacement (e.g., `$1`, `$2`). ### Matches that span inline tags are skipped Trados segments often contain inline tags – formatting marks, placeholders, field codes – that interrupt a run of plain text. If your search string would only match across one of these tag boundaries (for example, searching for `important thing` when the segment renders as `importantthing`), Replace and Replace All will **skip that segment** rather than apply a destructive flatten-and-rewrite that would lose the tag. You’ll see this in the status bar after a Replace All as `…, skipped N (match spans inline tags)`. The skipped segments are left untouched so the formatting survives; you can edit them manually if you want the replacement to happen. This applies to both the active-file path (Trados API replacements) and the on-disk path (SDLXLIFF XML rewrites). It only kicks in when the match genuinely straddles a tag – ordinary matches inside a single text run are replaced normally and tags are preserved. ## Regex Tips When the **.\*** checkbox is enabled, the search query is treated as a .NET regular expression. Some useful patterns: | Pattern | Matches | | ---------------- | --------------------------------------------------- | | `\bword\b` | ”word” as a whole word (not “keyword” or “wording”) | | `(word1\|word2)` | Either “word1” or “word2” | | `\d+` | One or more digits | | `"[^"]*"` | Anything inside double quotes | | `\s{2,}` | Two or more consecutive whitespace characters | Note Regex replace supports capture groups. For example, search for `(\w+)\s+(\w+)` and replace with `$2 $1` to swap two words. ## Keyboard Shortcuts | Shortcut | Action | | ----------------------------- | --------------------------------------------- | | **Alt+S** | Open SuperSearch (with selected text, if any) | | **Alt+W** | Search the web for the selected term | | **Enter** (in search box) | Start search | | **Enter** (in results grid) | Navigate to selected segment | | **Double-click** (result row) | Navigate to selected segment | ## Tips * Select a term in the editor and press **Alt+S** to instantly search for it across the entire project – or **Alt+W** to look it up on IATE, Linguee, Reverso and the rest, in your project’s language pair. * Use **Source only** scope to find segments where a particular term appears, then check how it was translated across files. * Use **Target only** scope with Replace to fix a consistent mistranslation across the entire project. * Use the **Files** and **TMs** buttons to limit the search to specific files or translation memories – useful in large projects where you only want to search a subset. * Switch the mode dropdown to **TMs** to use SuperSearch as a concordance tool, **Termbases** to search your terminology, or **Everything** to see project, TM and termbase hits side by side. * The status bar shows the number of results, what was searched, and the search time in milliseconds. * You can resize columns by dragging the column header borders. ## See Also * [Supervertaler](/trados/ai-assistant/) – AI-powered chat and context * [Batch Operations](/trados/batch-operations/) – Batch translate and proofread * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) – All shortcuts in one place # Support There are several ways to get help with Supervertaler for Trados. *** ## GitHub Discussions [GitHub Discussions](https://github.com/orgs/Supervertaler/discussions) is the main community hub for both Supervertaler for Trados and Supervertaler Workbench. It is the best place to: * Ask questions and get help from other users * Share tips, workflows, and best practices * Suggest features and improvements * Discuss anything related to Supervertaler Tip **[Visit Supervertaler Discussions →](https://github.com/orgs/Supervertaler/discussions)** *** ## Bugs & Feature Requests Use [GitHub Issues](https://github.com/Supervertaler/Supervertaler-for-Trados/issues) to report bugs or request new features. When reporting a bug, please include: * Your Trados Studio version * The Supervertaler plugin version * Steps to reproduce the problem * Any error messages or screenshots *** ## Email For private enquiries, contact . # Termbase Management Supervertaler for Trados uses the same SQLite termbase format as Supervertaler Workbench. You manage your termbases through the Settings dialogue. ## Accessing termbase settings 1. Click the **gear icon** in the TermLens panel, or go to **Settings** in the plugin ribbon 2. Switch to the **TermLens** tab ## Database file The plugin stores all termbases in a single `.db` file (SQLite database). * Click **Browse** to select an existing database file * Click **Create New** to create a fresh, empty database Note The `.db` file uses the same Supervertaler SQLite format as the standalone application. On Windows, you can share the same termbase file between both tools by pointing them to the same data folder. On a Mac with Parallels, see the note below. ## MultiTerm termbases If your Trados project has MultiTerm termbases (`.sdltb` files) attached, they appear automatically at the bottom of the termbase list with a **\[MultiTerm]** label and green background. These termbases are read-only in TermLens –to manage their terms, use Trados’s built-in MultiTerm interface. See [MultiTerm Support](/trados/multiterm-support/) for full details. ## Termbase list Once a database is loaded, the termbase list shows all Supervertaler termbases it contains, plus any detected MultiTerm termbases. Each Supervertaler termbase has three toggles: | Toggle | Purpose | | ----------- | ----------------------------------------------------------------------------------------------- | | **Read** | Load terms from this termbase for matching in TermLens | | **Write** | New terms added via [quick-add shortcuts](/trados/termlens/adding-terms/) go into this termbase | | **Project** | Designate as the project termbase (terms shown in pink, prioritised in matching) | Caution Only one termbase can be marked as **Project** at a time. Setting a new project termbase clears the flag from the previous one. ## Creating a new termbase 1. Click **Add Termbase** 2. Enter a **name** for the termbase 3. Select the **source language** and **target language** 4. Click **OK** The new termbase appears in the list, ready for use. ## Import from TSV You can import terminology from a tab-separated values file: 1. Select the target termbase in the list 2. Click **Import from TSV** 3. Select your `.tsv` file 4. A column-mapping dialog opens (from v18.20.187): one row per column in the file, with a sample of its contents and a dropdown saying which termbase field it goes to. It is pre-filled from the file’s headers, so a file exported by Supervertaler needs no changes – check the row count, termbase name and language pair in the heading, and click **Import**. For a file with no language headers, or with the target column first, set the source and target yourself; any column you do not want goes to *ignore* 5. A progress bar tracks the import (useful for large termbases with thousands of terms) If the file’s languages are the other way round from the termbase, or are not its pair at all, the dialog says so in a note under the grid. The mapping it suggests already accounts for a reversed file.  The column-mapping dialog, here on a Supervertaler export: every column recognised, the two language columns mapped to source and target. Each line is one column of the file; Row 1 to Row 3 show what that column holds. **File format:** The first row must be a header row. Headers the dialog recognises on its own (case-insensitive) – anything else can still be mapped by hand: | Column | Required | Recognised headers | | ---------- | -------- | -------------------------------------------------------------------- | | **Source** | Yes | `Source`, `Source Term`, `Src`, or a language name (e.g., `English`) | | **Target** | Yes | `Target`, `Target Term`, `Tgt`, or a language name (e.g., `Dutch`) | | Term UUID | No | `Term UUID`, `UUID`, `Term ID`, `ID` | | Priority | No | `Priority`, `Prio`, `Rank` | | Domain | No | `Domain`, `Subject`, `Field`, `Category` | | Definition | No | `Definition` | | Notes | No | `Notes`, `Note`, `Comment`, `Description` | | Project | No | `Project` | | Client | No | `Client`, `Customer` | | Forbidden | No | `Forbidden`, `Do not use` | For terms with multiple synonyms, use pipe-delimited values: `main|synonym1|synonym2`. Forbidden synonyms are wrapped as `[!term]`. **Example:** ```plaintext Source Target Domain Notes database databank|gegevensbank software software Non-translatable user interface gebruikersinterface|gebruikersomgeving IT ``` Note TSV files exported from Supervertaler (both the Trados plugin and Workbench) can always be reimported without any changes. Files from other tools are also supported as long as they have recognisable column headers. ## Export To export all terms from a termbase: 1. Select the termbase in the list 2. Click **Export** 3. Pick the format in the save dialog, then choose a location Three formats, for three different purposes. ### TSV – the one that comes back unchanged Tab-separated columns with a header row: `Term UUID`, `Source`, `Target`, `Priority`, `Domain`, `Notes`, `Project`, `Client`, `Forbidden`. Synonyms are pipe-delimited and forbidden synonyms are marked with `[!term]`. UTF-8 with BOM, for Excel compatibility. Use it for a backup, for editing in a spreadsheet, or for moving terms between Supervertaler termbases – it reimports here with nothing lost. Note Terms whose own text contains a `|` or a `\` are escaped from v18.20.177. A TSV written by an **earlier** build cannot be repaired: in it, a delimiter and a literal pipe are the same character. If you have such a file and the terms matter, re-export it. ### MultiTerm XML – for Trados The format MultiTerm and Glossary Converter import. This is how you get terms *out* of Supervertaler and into a Trados termbase: * **For a `.sdltb`**: convert the XML with Glossary Converter, or import it in MultiTerm. * **For a `.ttb`** (Studio 2026): in Studio’s **Termbases** view, create a termbase and use **Import Terms**. Carries source and target terms with their synonyms, plus definition, domain, notes, context, part of speech, URL, client, project and the forbidden flag – more than the TSV export, which has no column for several of those. ### TBX – for everything else TBX-Basic (ISO 30042), the standard interchange format. MultiTerm reads it and so do most other CAT tools, so it is the better choice if your terminology has to travel beyond Trados. Note **Supervertaler cannot write a `.sdltb` or `.ttb` directly.** Those are a Microsoft Access database and an undocumented SQLite format respectively; producing one means guessing at a file layout that is not published, and a termbase Studio only half-accepts would be worse than one it refuses outright. Both formats above are documented and Trados imports them, at the cost of one conversion step you perform yourself. ### What a round trip keeps, and what it doesn’t Terms, synonyms and the descriptive fields survive a full circle out to a Trados termbase and back. The *structure* does not: a MultiTerm entry is concept-oriented and can hold many languages, while a Supervertaler termbase is bilingual rows – so one conversion handles one language pair, and concepts flatten to pairs with any extra terms in a language becoming synonyms. ## Import from a Trados termbase **Import .sdltb/.ttb…** copies terms *out of* a Trados termbase and *into* a Supervertaler one. You choose the language pair, whether to create a new Supervertaler termbase or add to an existing one, and which of the Trados descriptive fields map to definition, domain, notes and so on. The Trados termbase is only ever read, never modified. An AI assistant connected through the [MCP server](/trados/mcp-server/) can do the same in one step with `import_project_termbase`, including a dry run that reports what would be imported before anything is written. ## Termbase Editor For full editing capabilities, double-click a termbase in the list to open the **Termbase Editor**. From here you can: * **Search** for terms by source or target text * **Edit** individual term entries * **Delete** terms * Perform **bulk operations** (e.g. bulk delete, bulk reverse) ### Right-click menu Right-clicking any row in the grid opens a context menu with the following actions: * **Copy cell** – copies the content of the clicked cell to the clipboard. * **Edit term…** – opens the full term entry editor for the clicked row. * **Reverse source/target** – swaps the source and target for the selected rows (see below). * **Delete term** – deletes the selected rows after confirmation. Multi-row selection is preserved: if you select several rows first and then right-click on one of them, the selection stays intact so actions apply to all selected entries. If you right-click on a row that wasn’t already selected, the selection collapses to just that row. ### Reversing source/target If you have term entries that ended up in the wrong direction – for example, English text in the Dutch column when the termbase is declared English → Dutch – you can correct them with **Reverse source/target**: 1. Select one or more rows in the grid (Shift-click or Ctrl-click for multi-select). 2. Right-click → **Reverse source/target (N entries)**. 3. Confirm. The operation swaps the source and target text, language tags, abbreviations, and flips the direction of every linked synonym. It runs in a single database transaction, so a partial failure leaves the termbase untouched. This action is mostly for repairing legacy entries created or edited under v4.19.24 or earlier, when the term entry editor could write values into the wrong DB columns in projects whose direction was the inverse of the termbase’s. From v4.19.25 onwards the editor guards against that, so new entries should not need this repair. Note **Add and Edit dialog fields are always in termbase direction.** The dialog labels and values both reflect the termbase’s declared direction – English on the left when the termbase is declared EN→NL, regardless of the current Trados project’s direction. From v4.19.25 the values are guaranteed to align with the labels: the Edit dialog re-reads the entry from the database, and the Add dialog swaps the pre-fills internally when the project direction is the inverse of the termbase. Earlier versions could silently write reversed entries in inverse-direction projects – use **Reverse source/target** above to repair any pre-v4.19.25 damage. ## Sharing termbases Tip **Tip:** Keep the `.db` file on a network drive or cloud-synced folder (OneDrive, Dropbox, Google Drive) to share termbases across machines and with colleagues. Since both the Trados plugin and Supervertaler Workbench use the same format, everyone can work from the same terminology. Caution **Mac users (Parallels):** On a Mac, Supervertaler Workbench runs natively on macOS while the Trados plugin runs inside Parallels (Windows). The two products cannot share the same `.db` file directly because the Trados plugin must store its data on the Windows side (`C:\Users\...`) – not on the Mac-side shared folder (`\\Mac\Home\...`). To keep your termbases in sync, export from one side and import on the other after making changes. This is a limitation of Parallels’ virtual network filesystem, not of the termbase format itself. *** ## See Also * [MultiTerm Support](/trados/multiterm-support/) * [TermLens Settings](/trados/settings/termlens/) * [Adding & Editing Terms](/trados/termlens/adding-terms/) * [Data Folder](/trados/data-folder/) # TermLens TermLens is an inline terminology display that shows the source text of the current segment word by word, with termbase translations directly underneath each matched term. It updates automatically when you navigate to a new segment.  ### How It Works When you select a segment in the Trados editor, TermLens analyses the source text against all active termbases and displays the result in a visual layout: * **Matched words** appear with their termbase translation underneath, on a coloured background * **Unmatched words** are shown in light grey text so you can read the full source sentence in context This gives you an at-a-glance overview of every term in the segment that has a termbase entry – without hovering or clicking anything. #### Selection tracking Selecting text in the editor’s **source segment** highlights the corresponding words in TermLens with a soft yellow band – matched and unmatched words alike. On a long segment this anchors your eye to the part you are actually working on, with the relevant term chips right there. The highlight follows your selection live and clears when you deselect. When the same phrase occurs more than once in a segment, the occurrence nearest your cursor is highlighted.  Select text in the source segment and TermLens highlights the same words – your eye lands straight on the relevant term chips. Selecting text in the **target segment** works in reverse: the term chips whose translation your selection covers light up, including matches via target synonyms and abbreviations. Only chips can light up from a target selection – the editor provides no word alignment between source and target, so unmatched words have no reliable counterpart to point at. Whichever side you selected last drives the highlight. ![]() ### Colour Coding TermLens uses five background colours to distinguish term types: | Colour | Meaning | | ---------- | ----------------------------------------------------------------------- | | **Blue** | Regular Supervertaler termbase match | | **Purple** | Abbreviation match (matched via source abbreviation, not the full term) | | **Pink** | Project termbase match (higher priority) | | **Yellow** | Non-translatable term (source = target) | | **Green** | MultiTerm termbase match (`.sdltb`) | Note Designate one termbase as the **Project termbase** in settings to make its terms appear in pink. Project terms take visual priority over regular terms, making it easy to spot client-specific terminology. Tip **MultiTerm termbases** attached to your Trados project appear automatically as green chips. They are read-only – to edit MultiTerm terms, use Trados’s built-in MultiTerm interface. See [MultiTerm Support](/trados/multiterm-support/) for details. ### Chip Indicators In addition to colour coding, TermLens shows small indicators in the top-right corner of term chips: | Indicator | Meaning | | -------------- | ----------------------------------------------------------------------------------- | | **≡** (indigo) | The entry has synonyms (source-side, target-side, or both). Hover to see them. | | **●** (amber) | The entry has metadata – a definition, domain, notes, or URL. Hover to see details. | Both indicators can appear simultaneously. Hover over any term chip to see an interactive popup with full details, including source synonyms (prefixed with “Also:”), target synonyms (shown as bullet points), definitions, domain, notes, and clickable URLs. The popup stays open when you move the mouse into it, so you can click on links. #### Markdown Rendering The **Notes** and **Definition** fields in the term popup support Markdown formatting. If the content contains Markdown syntax (tables, bold, italic, headings, bullet lists, code blocks), it is rendered with proper formatting instead of plain text. This is especially useful when AI-generated term notes include structured data like translation tables. ![]() Markdown formatting rendered in TermLens term popup #### Resizable Popup You can resize the term popup by dragging the grip in the bottom-right corner. The width is remembered for the rest of the session, so subsequent popups open at your preferred size. ### Inserting Terms #### Click to Insert Click any translation shown under a source word. The translation is inserted at the cursor position in the target field. #### Automatic capitalisation Displayed and inserted terms follow the capitalisation of the source occurrence in the segment, not the capitalisation stored in the termbase. A term stored as “More preferably” shows and inserts as “more preferably” when the segment contains it lower-case mid-sentence; a lower-case stored term is capitalised when the occurrence starts the sentence; and an ALL-CAPS occurrence (a heading, say) upper-cases the whole term. This applies to every insertion path – chip clicks, Alt+digit shortcuts, the TermLens popup and TermPicker. The rules are deliberately conservative: acronyms and mixed-case terms (MRI, pH) are never altered, and abbreviation matches keep their stored casing. You can switch the behaviour off with **Adapt term capitalisation to the segment** in [TermLens settings](/trados/settings/termlens/#adapt-term-capitalisation). #### Keyboard Shortcuts (Alt+1 through Alt+9) Each matched term in TermLens is assigned a **numbered badge**. Press **Alt+1** to insert the first match, **Alt+2** for the second, and so on up to **Alt+9**. #### Shortcuts for Terms 10+ For terms numbered 10 and above, TermLens supports two shortcut styles (configurable in Settings): * **Sequential** (default) – type the term number digit by digit: `Alt+14` inserts term 14 * **Repeated digit** – press the same digit key multiple times: `Alt+55` inserts term 14 (5th term in the second tier: 9+5) The badge on each term chip shows exactly which key combination to use. See [Keyboard Shortcuts](/trados/keyboard-shortcuts/) for details on both modes. Note Terms beyond 45 have no keyboard shortcut. Use the **TermLens popup** or **TermPicker** to insert them. #### TermLens popup (Alt+L) Press **Alt+L** to open the [**TermLens popup**](/trados/termlens/termlens-popup/) – a borderless floating version of this panel for the active segment. Designed for keyboard-only term selection on small screens where keeping the docked panel always-visible costs too much vertical space. Press **Alt+L** again to close. Inside, **Right / Down / Tab** cycle the highlighted match, **Enter** inserts and closes, **Escape** dismisses without inserting. #### TermPicker (Alt+P) For segments with many matches and when a sortable list view is preferable, press **Alt+P** to open [**TermPicker**](/trados/termlens/termpicker/). It shows all matched terms in a list – synonyms expanded – and lets you insert any term with a double-click or Enter. TermPicker is a sibling surface to TermLens: same termbase data, different ergonomics (flat list vs in-context chips). TermPicker is also available as a **dockable pane** (Studio’s **View** tab → **TermPicker**), so you can keep the list permanently visible instead of the TermLens panel if that suits you better. Both views are available in both placements – docked or as a popup at the cursor.  ### Right-Click Context Menu Right-click any term in TermLens to access: | Action | Description | | ---------------------------- | ---------------------------------------------------------------------------------- | | **Edit Term** | Open the term editor to modify source, target, or metadata | | **Delete Term** | Remove the term from the termbase | | **Mark as Non-Translatable** | Flag the term so it appears in yellow (source = target) | | **Mark as Translatable** | Remove the non-translatable flag (shown when the term is already non-translatable) | ### Quick-Add Terms You can add terms without opening a dialogueue: | Shortcut | Action | | -------------- | ---------------------------------------------------------------------------------------- | | **Alt+Down** | Quick-add the selected text to all write termbases | | **Alt+Up** | Quick-add the selected text to the project termbase | | **Ctrl+Alt+T** | Open the Add Term Entry dialogue (full editor: definition, domain, notes, URL, synonyms) | | **Ctrl+Alt+N** | Quick-add the selected text as a non-translatable term | Tip Quick-add shortcuts use the currently selected source text and the corresponding selected or clipboard target text. The term is added instantly without opening a dialogue. ### Font Size Use the **A+** and **A-** buttons in the TermLens panel header to increase or decrease the font size. Changes apply immediately. ### Tips * TermLens respects termbase activation –only terms from activated termbases are shown. * If you have many termbases, designate one as the **Project termbase** (shown in pink) to make its terms stand out. * Hover over a term to see an interactive popup with all translations, synonyms, abbreviation pairs, definitions, URLs, and the termbase name. The popup stays open when you move the mouse into it, allowing you to click on URLs. * A small **indigo ≡ indicator** appears in the top-right corner of a term chip when the entry has synonyms. An **amber dot** appears when the entry has metadata (definition, domain, notes, or URL). Both indicators can appear together. * If a term has an **abbreviation** (e.g., “GC” for “gaschromatografie”), both the full term and the abbreviation are highlighted when they appear in the same segment. The abbreviation chip shows the abbreviated translation; the full-term chip shows the full translation. *** ### See Also * [Adding & Editing Terms](/trados/termlens/adding-terms/) * [TermPicker](/trados/termlens/termpicker/) * [MultiTerm Support](/trados/multiterm-support/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) * [Getting Started](/trados/getting-started/) # Adding Terms Supervertaler for Trados provides several ways to add, edit, and manage terminology without leaving the Trados editor. ## Quick-add (Alt+Down) The fastest way to add a term while translating: 1. Select the **source text** you want to add as a term 2. Select the **target text** (the translation) 3. Press **Alt+Down** The term is added instantly to all **write-enabled** termbases. No dialogue, no interruption. Note Quick-add writes to every termbase that has **Write** enabled in your [TermLens Settings](/trados/settings/termlens/). If you want to target a specific termbase, use the Add Term Entry dialogue (Ctrl+Alt+T) instead. ## Quick-add to project termbase (Alt+Up) Works the same as Alt+Down, but adds the term specifically to the **project termbase** (the termbase marked as “Project” in settings). Use this when you want to keep client-specific terminology separate and prioritised. 1. Select the **source text** 2. Select the **target text** 3. Press **Alt+Up** ## Quick-add non-translatable (Ctrl+Alt+N) For terms that should remain identical in source and target (brand names, product codes, abbreviations): 1. Select the text in the **source** field 2. Press **Ctrl+Alt+N** This creates a term entry where source and target are the same. Non-translatable terms appear as **yellow** chips in [TermLens](/trados/termlens/), the [TermLens popup](/trados/termlens/termlens-popup/) and [TermPicker](/trados/termlens/termpicker/). ## Add term with abbreviation (Ctrl+Alt+A) When a segment spells a concept out in full **and** gives its abbreviation, this fills the dialogue in for you. Take a segment like: > Deze verklaring wordt opgesteld conform de **Sustainable Finance Disclosure Regulation** (**SFDR**, Verordening (EU) 2019/2088). Press **Ctrl+Alt+A** (or right-click → **Add term with abbreviation (AI)**) and the **Add term entry** dialogue opens with the source term, the target term and both **Abbreviation** fields already filled in. Check them and click **Add**. 1. Optionally select any part of the term you want 2. Press **Ctrl+Alt+A** 3. Review the pre-filled dialogue and click **Add** **Your selection decides which term.** Select any part of a term – or a whole phrase containing it – and that is the term you get, completed to its full extent. The AI works out where the term begins and ends and finds its abbreviation; it will not swap in a different term. With nothing selected, it looks at the whole segment. Note This handles **only** terms that carry an abbreviation. If the segment has none, the ordinary dialogue opens with your selection instead, exactly as Ctrl+Alt+T would. For terms without an abbreviation, use Alt+Down or Ctrl+Alt+T, which add precisely what you selected. Caution Nothing is saved until you click **Add**. Do check the fields – the AI is restricted to copying text that genuinely appears in the segment and can never invent an abbreviation, but it can still pick a wider or narrower span than you had in mind. Requires an AI provider configured in [AI Settings](/trados/settings/ai-settings/); it uses whichever provider and model you have selected. ## Add term entry (Ctrl+Alt+T) For full control over a new term, press **Ctrl+Alt+T** (or right-click in the editor and choose **Add Term…**). This opens the **Add term entry** dialogue, which lets you fill in all fields before saving: | Field | Description | | ----------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Source** | The source-language term | | **Target** | The target-language translation | | **Source Abbreviation** | Optional abbreviated form (e.g. “GC” for “gaschromatografie”). Separate multiple variants with `\|`. | | **Target Abbreviation** | Optional abbreviated form of the target term | | **Source synonyms** | Alternative source-language forms for the same concept | | **Target synonyms** | Alternative target-language translations | | **Definition** | Optional definition or usage note. Supports multiple lines – click the **▼** button to expand the field for longer content. | | **Domain** | Subject area (e.g. “Legal”, “Patents”, “Medical”) | | **Notes** | Any additional notes for translators. Supports multiple lines with an expand button, like Definition. | | **URL** | Optional reference URL (shown as a clickable link in the term popup) | | **Client** | Optional client code (e.g. “ACME”, “GLOBEX”). A label for your own filtering and reporting. It no longer selects SuperMemory content: the bank you pick in the toolbar is what the AI reads. | | **Project** | Optional project name (e.g. a job code or client-side project ID). Bookkeeping field for the user’s own organisation – not sent to the AI in translation prompts. The Termbase Editor’s grid lets you sort and filter by Project. | | **Non-translatable** | Check this to mark the term as non-translatable | The term is added to the **project termbase** if one is configured, or the first write-enabled termbase otherwise. Caution **Trados conflict:** Trados Studio assigns **Ctrl+Alt+T** to “Insert TM Symbol” by default. If pressing Ctrl+Alt+T does nothing, you need to remove Trados’s binding first. Go to **File → Options → Keyboard Shortcuts**, search for “Insert TM Symbol”, and delete or reassign its shortcut. Then Ctrl+Alt+T will work as expected in Supervertaler. Tip **Tip:** Use the project termbase for client-specific terminology that should be prioritised over background termbases. Project termbase terms appear in pink in TermLens. ## Smart selection You don’t need to precisely select entire words when adding terms. All quick-add shortcuts (**Alt+Down**, **Alt+Up**, **Ctrl+Alt+N**, **Ctrl+Alt+T**) automatically expand your selection to the nearest word boundaries. For example, to add **standalone version** = **zelfstandige versie** to your termbase, it’s enough to select **alone ver** in the source and **andige ver** in the target. Supervertaler expands both selections to the full words automatically. This means you can work fast and loose with your mouse or keyboard selections – no need for the precise click-and-drag that normally slows you down. Just grab roughly the right area and Supervertaler takes care of the rest. ### How it works When you make a selection, Supervertaler scans the full segment text for every occurrence of your selected text and applies these rules, in order: 1. **Exact word match wins** – if the selection matches a complete word somewhere in the segment (i.e. it sits between spaces or punctuation), that word is used as-is. For example, if the segment contains both *hechtingsbevorderaars* and *hechting*, selecting **hechting** returns **hechting** – the exact word – not the longer compound. 2. **Shortest word wins** – if the selection is embedded inside multiple words, the shortest enclosing word is preferred. For example, if the segment contains *hechtingsbevorderaars* and *hechting*, selecting **echt** returns **hechting** (8 characters) rather than *hechtingsbevorderaars* (21 characters), because the user most likely intended the simpler word. 3. **Single match expands** – if the selection appears inside only one word, it expands to that word’s boundaries. ### Tips for reliable results * **Select at least 3–4 characters** – very short selections (1–2 characters) may match common short words elsewhere in the segment (e.g., selecting **he** could match the word *the*) * **Select the whole word when in doubt** – if a segment contains similar-looking words and you want a specific one, a complete-word selection is always matched correctly * **Use Ctrl+Alt+T for tricky cases** – the Add Term Entry dialogue lets you review and edit the expanded term before saving, so you can catch any unexpected expansion Note Press **F2** to manually expand your current selection to word boundaries without adding a term. This lets you preview exactly what Supervertaler would capture before pressing a quick-add shortcut. ## Merge prompt When you add a term and the **source** or **target** already exists in the termbase (but with a different translation), Supervertaler shows a prompt asking what you want to do: * **Add as Synonym** – merges the new translation into the existing entry as a synonym, keeping your termbase tidy * **Add & Edit…** – adds the synonym and opens the Term Entry Editor so you can review the metadata before saving * **Keep Both** – creates a separate entry alongside the existing one * **Cancel** – aborts the operation The merge prompt always displays terms in your **project’s language direction**, regardless of how the termbase stores them internally. For example, in a Dutch → English project using an English → Dutch termbase, the dialogue shows the Dutch source term first and the English target term second. **Example:** Your termbase already has **adhesion → hechting**. You select **adhesion → aanhechting** and press Alt+Down. The merge prompt appears because the source term “adhesion” already exists. Clicking “Add as Synonym” adds *aanhechting* as a target synonym of the existing entry, so both translations are grouped together. Note The merge prompt only appears when the source or target matches exactly (case-insensitive). It does **not** apply to non-translatable quick-add (**Ctrl+Alt+N**). ## Editing existing terms To edit a term that already exists in your termbase: 1. Right-click the term in the **TermLens** panel 2. Select **Edit Term…** 3. The **Term Entry Editor** opens, where you can: * Modify the source or target text * Add or remove **synonyms** (multiple translations for one source term) * Update the definition * Toggle the non-translatable flag Click **Save** when done. ## Abbreviations Term entries can have optional **source and target abbreviation** fields. When a source abbreviation appears in a segment, TermLens highlights it and shows the target abbreviation underneath – just like a regular term match. ### Adding abbreviations 1. Open the **Term Entry Editor** (right-click a term → Edit Term) 2. Fill in the **Source Abbreviation** and **Target Abbreviation** fields 3. Click **Save** ### Multiple abbreviation variants You can specify multiple variants of the same abbreviation by separating them with a **pipe character** (`|`): ```plaintext GC|G.C.|gc|g.c. ``` Each **source** variant is indexed and matched independently, so all common forms of the abbreviation are recognised in the source text. The **first variant** is used as the display text and for insertion. Note The two sides behave differently, so put your variants in the **Source Abbreviation** field. Every source spelling is matched wherever it appears in the segment, but only the **first** target variant is ever used – it is the form inserted into your translation and named in AI prompts. Additional target variants are stored but have no effect. ### How abbreviation matching works When both the full term and its abbreviation appear in the same segment (e.g., “gaschromatografie (GC)”), TermLens shows **both** as highlighted chips: * The **full term** chip shows the full target translation (e.g., “gas chromatography”) – displayed in the regular **blue** colour * The **abbreviation** chip shows the target abbreviation (e.g., “GC”) – displayed in **lavender** so it is instantly distinguishable from a full-term match Clicking or Alt+digit-inserting an abbreviation chip inserts the **target abbreviation** (first variant), not the full target term. Note Abbreviations are also included in AI translation prompts, so the AI knows both the full term and its abbreviated form. ## Deleting terms 1. Right-click the term in the **TermLens** panel 2. Select **Delete Term** 3. Confirm the deletion in the dialogue Caution Deletion is permanent. The term is removed from the termbase database file. ## Bulk Add Non-Translatable For adding many non-translatable terms at once (e.g., a list of brand names or product codes): 1. Open **Settings** (gear icon in the TermLens panel) 2. Find the **Bulk Add Non-Translatable** option 3. Paste your terms, **one per line** 4. Click **Add** to save them all at once *** ## See Also * [TermPicker](/trados/termlens/termpicker/) * [Termbase Management](/trados/termbase-management/) * [TermLens Settings](/trados/settings/termlens/) # TermLens popup The **TermLens popup** is a borderless floating version of the docked TermLens panel for the active segment. Designed for keyboard-only term selection on small screens – and for translators who want to insert terms without ever reaching for the mouse.  The TermLens popup with the current match highlighted (amber ring on the source word) ### When to use it * **Small screens / laptops** – keeping the docked TermLens panel always-visible can cost too much vertical space, especially for longer source sentences. The popup gives you the same view on demand and disappears when you’re done. * **Pure-keyboard workflows** – Alt+L, cycle, Enter, back to typing. No mouse, no menu hunting. ### Opening and closing | Key | Action | | ----------------------- | ------------------------------------------------ | | **Alt+L** | Toggle the popup (open if closed, close if open) | | **Escape** | Close without inserting | | Click outside the popup | Close without inserting | You can change the key in **File → Options → Keyboard Shortcuts**. (Earlier versions listed **Ctrl+Alt+G**; that key now belongs to [AutoTagger](/trados/autotagger/).) **Was a Ctrl tap until this version.** Pressing and releasing Ctrl on its own used to open the popup – a quick gesture, but not one Studio could tell apart from any other program’s Ctrl-modified shortcut once something else had consumed the middle key. The popup opened by itself for anyone running a keyboard tool, a text expander or voice commands alongside Studio. Alt+L is unambiguous. If you preferred the tap, note that Studio remembers per-user shortcuts, so an existing installation keeps whatever you already had bound. ### Cycling between matches When the popup opens, the first match has an amber ring around its source word – that is the **current match** that Enter will insert. | Key | Action | | --------------------------------- | -------------------------------------------------- | | **Right** / **Down** / **Tab** | Move the current-match highlight to the next match | | **Left** / **Up** / **Shift+Tab** | Move it to the previous match | Cycling wraps: from the last match, Right takes you back to the first. ### Inserting | Key / action | Result | | ------------------ | -------------------------------------------------------------------------------------------------- | | **Enter** | Insert the current match into the target segment, close the popup, return focus to the target cell | | **Click any chip** | Insert that match into the target segment, close the popup, return focus to the target cell | Both paths share the same insertion logic – there is no difference between picking by keyboard and picking by mouse. ### Editing a match Press **E** while a match is highlighted to open the term-entry editor for that entry. The popup closes first so the editor opens with clean focus. The editor is the same dialogue the docked panel’s right-click “Edit Term…” menu uses, including the multi-termbase editing case for entries that exist in more than one termbase. Note **MultiTerm matches are read-only** in TermLens. Pressing E on a green MultiTerm chip flashes a hint instead – edit those entries in **Trados → Termbase Viewer**. ### Visuals The popup uses the same chip rendering, colour scheme, and metadata indicators as the docked TermLens panel – pink for project termbase terms, blue for regular Supervertaler terms, yellow for non-translatable, green for MultiTerm. See the [TermLens overview](/trados/termlens/) for the full colour key. ### TermLens popup vs TermPicker Both show the same matches for the active segment. Pick whichever fits your style: | | TermLens popup (Alt+L) | [TermPicker](/trados/termlens/termpicker/) (Alt+P, or docked) | | -------- | ------------------------------------------------------ | ------------------------------------------------------------- | | Layout | Source segment with chips underneath each matched word | Sortable, scrollable table | | Best for | Skimming matches in segment context | Many matches that benefit from sorting / typing-to-jump | | Keyboard | Arrow / Tab cycles a highlighted match | 0–9 jumps directly; Up / Down navigate | | Modality | Modeless – click outside to dismiss | Modal – Escape or Cancel to close | *** ### See Also * [TermLens overview](/trados/termlens/) * [TermPicker](/trados/termlens/termpicker/) * [Keyboard shortcuts](/trados/keyboard-shortcuts/) # TermPicker **TermPicker** shows all matched terms for the current segment as a sortable, keyboard-navigable list. It is useful when [TermLens](/trados/termlens/) shows many matches and you want a quick overview without scrolling – and it is available either as a popup at your cursor or as a permanently docked pane.  TermPicker with all matched terms for the current segment ### Two views, two placements TermLens and TermPicker are sibling surfaces: both show the terminology matches for the current segment, drawn from the same termbases, but they present them differently. Each is available in both placements, so you can choose the **view** you prefer independently of **where** you want it: | | Docked pane | Popup at the cursor | | -------------- | --------------- | ------------------- | | **TermLens** | TermLens panel | tap **Ctrl** | | **TermPicker** | TermPicker pane | **Alt+P** | * **TermLens** shows matches *in context* – the source sentence with terms highlighted in place. Best for reading and scanning. * **TermPicker** shows the same matches as a flat sortable list. Best for quick insertion and for seeing every synonym at once. ### Opening TermPicker **As a popup:** press **Alt+P**. It appears above the editor with every synonym group already expanded, so you can see all alternatives at a glance. **Escape** closes it. **As a docked pane:** open it from Studio’s **View** tab → **TermPicker**. The pane stays visible and updates as you move through the document, in step with the TermLens panel. Drag it wherever suits you – Studio remembers the position. A layout that works well is the Translation Results window at the top right with the TermPicker pane directly below it. Note When the pane is open, **Alt+P moves the keyboard focus into it** rather than opening the popup on top of it – so the same shortcut always gets you to the list, wherever you keep it. With no pane in your layout, Alt+P opens the popup as usual. ### Colour-coded rows Each row in TermPicker is colour-coded by termbase type: | Colour | Meaning | | ---------- | ----------------------------------- | | **Pink** | Project termbase term | | **Blue** | Regular Supervertaler termbase term | | **Yellow** | Non-translatable term | | **Green** | MultiTerm termbase term (`.sdltb`) | This lets you instantly see where each term comes from and how it should be handled. ### The details dot An **amber dot** in the second column marks terms that carry extra information – a definition, domain, notes or a URL. Press **I** on such a row to see it (see below). Rows without a dot have nothing further to show, so you never have to guess. This mirrors the dot on the [TermLens](/trados/termlens/) chips. ### Expandable synonyms Terms with multiple translations open **already expanded**, with their alternatives listed as sub-rows beneath the term, so nothing is hidden behind a marker. Press **Left arrow** to collapse a group (the indicator changes to **▸**) and **Right arrow** to expand it again. ### Keyboard navigation TermPicker is designed for fast keyboard use, and the keys are identical in the popup and the pane: | Key | Action | | ------------- | ------------------------------------------------------------------ | | **Up / Down** | Navigate between rows (wraps around) | | **Right** | Expand synonyms for the selected term | | **Left** | Collapse synonyms | | **0-9** | Jump directly to that term number | | **I** | Show the term’s details – definition, domain, notes, URL, synonyms | | **E** | Open the term editor for the selected term | | **Enter** | Insert the selected term | | **Escape** | Close the popup without inserting | Navigation wraps around: pressing **Down** on the last row jumps to the first, and **Up** on the first jumps to the last. Note **I** and **E** work the same way here as in the [TermLens popup](/trados/termlens/termlens-popup/). Press **I** again to dismiss the details popup. MultiTerm terms cannot be edited – those termbases are read-only in Supervertaler; edit them in Trados’s own MultiTerm interface. ### Term details (I) Pressing **I** on a row opens a small window just below it showing everything the termbase holds for that term: its synonyms, definition, domain, notes and URL, for every entry behind the row. Press **I** again – or **Escape** – to dismiss it. It also closes by itself when you move to another row, so it can never describe the wrong term. ### Right-click menu Right-clicking a row gives you the same three actions as right-clicking a TermLens chip: | Action | Effect | | ---------------------------- | -------------------------------------------------------- | | **Edit Term…** | Opens the term editor (also **E**) | | **Mark as Non-Translatable** | Flags the term so source = target (or removes the flag) | | **Delete Term** | Removes the term from the termbase, after a confirmation | MultiTerm terms are read-only in Supervertaler, so these actions are greyed out for them – edit those in Trados’s own MultiTerm interface. ### Inserting a term You can insert a term in two ways: * **Double-click** any row to insert that term at the cursor position in the target field * **Press Enter** on the selected row From the popup, inserting also closes it; from the docked pane, the list stays open so you can insert several terms in a row. Like the TermLens chips, the picker displays and inserts terms with the capitalisation of the segment occurrence (the first occurrence when a term appears more than once) rather than the stored capitalisation – see [Adapt term capitalisation](/trados/settings/termlens/#adapt-term-capitalisation). *** ### See Also * [TermLens](/trados/termlens/) * [TermLens popup](/trados/termlens/termlens-popup/) * [Adding & Editing Terms](/trados/termlens/adding-terms/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) * [Termbase Management](/trados/termbase-management/) # Replace curly quotes with straight quotes Text transforms are a special type of QuickLauncher prompt that performs local find-and-replace operations on the active target segment – instantly, without calling an AI provider. ## When to use text transforms Text transforms are useful for cleaning up invisible or problematic characters in your translations. For example: * **InDesign (IDML) forced line breaks** – InDesign uses invisible Unicode LINE SEPARATOR (U+2028) characters as forced line breaks (Shift+Enter). These are invisible in Trados but can cause problems in your translations. * **Zero-width spaces and joiners** – invisible characters from PDF or web sources * **Normalising quotes or dashes** – replacing curly quotes with straight quotes, or em dashes with en dashes ## How it works 1. Navigate to the segment you want to clean 2. Press **Alt+Q** (or right-click → QuickLauncher) 3. Open the **Text operations** folder 4. Click **Strip U+2028** (or another transform) The transform runs instantly. A dialogue confirms how many replacements were made, and the cleaned text is copied to your clipboard. Note Text transforms modify the **target segment** only. The source segment is never changed. ## Built-in transforms Supervertaler ships with one built-in text transform: ### Strip U+2028 Removes invisible Unicode LINE SEPARATOR (U+2028) and PARAGRAPH SEPARATOR (U+2029) characters from the target segment, replacing them with spaces. Consecutive spaces are collapsed to a single space. These characters are commonly inserted by InDesign (IDML) as forced line breaks (Shift+Enter). They are invisible in the Trados editor but can corrupt translations – the AI may produce spurious line breaks, or the characters may cause formatting issues in the final document. ## Creating your own transforms Text transforms are stored as `.md` files in your prompt library, just like regular prompts. The only difference is the YAML frontmatter has `type: transform` instead of `type: prompt`, and the content body contains find/replace rules instead of a prompt. ### Step by step 1. Open **Settings → Prompts** 2. Click **New** 3. Set the **Category** to `QuickLauncher/Text operations` (or any QuickLauncher subfolder) 4. In the YAML frontmatter, change `type: prompt` to `type: transform` 5. In the content body, write your find/replace rules ### Rule format Each rule is a `find:` / `replace:` pair. Blank lines between rules are optional but improve readability. Lines starting with `#` are comments. ```plaintext find: "\u201C" replace: "\u0022" find: "\u201D" replace: "\u0022" ``` ### Unicode escapes Use `\uXXXX` to specify Unicode characters by their code point. This is essential for invisible characters that cannot be typed or seen in a text editor. | Escape | Character | Description | | -------- | ----------- | ------------------------------------------- | | `\u2028` | (invisible) | LINE SEPARATOR – InDesign forced line break | | `\u2029` | (invisible) | PARAGRAPH SEPARATOR | | `\u200B` | (invisible) | ZERO WIDTH SPACE | | `\u200C` | (invisible) | ZERO WIDTH NON-JOINER | | `\u200D` | (invisible) | ZERO WIDTH JOINER | | `\uFEFF` | (invisible) | BYTE ORDER MARK (BOM) | | `\u00A0` | (invisible) | NON-BREAKING SPACE | | `\u201C` | ” | LEFT DOUBLE QUOTATION MARK | | `\u201D` | ” | RIGHT DOUBLE QUOTATION MARK | ### Example: Strip U+2028 (built-in) ```yaml --- type: transform name: "Strip U+2028" description: "Removes invisible Unicode LINE SEPARATOR and PARAGRAPH SEPARATOR" category: "QuickLauncher/Text operations" default: true --- # Strip invisible Unicode line/paragraph separators. # These are commonly inserted by InDesign (IDML) as forced line breaks. find: "\u2028" replace: " " find: "\u2029" replace: " " ``` ### Example: Normalise non-breaking spaces ```yaml --- type: transform name: "Fix non-breaking spaces" description: "Replaces non-breaking spaces with regular spaces" category: "QuickLauncher/Text operations" --- # Replace non-breaking spaces (U+00A0) with regular spaces find: "\u00A0" replace: " " ``` ## Keyboard shortcuts Text transforms appear in the QuickLauncher menu alongside regular prompts. You can assign them to keyboard slots (Ctrl+Alt+1 through Ctrl+Alt+0) for instant access: 1. Open **Settings → Prompts** 2. Select the transform in the tree 3. Choose a **Shortcut** slot from the dropdown at the bottom ## Clipboard After a transform runs, the cleaned target text is automatically copied to your clipboard. This is useful if you need to paste the cleaned text elsewhere – for example, into a text editor or a QA tool. ## Technical notes * Transforms use Trados’s `ProcessSegmentPair` API to commit changes, the same mechanism used by Batch Translate. This ensures all formatting tags (bold, italic, etc.) are preserved. * After replacements, consecutive spaces are collapsed to a single space to prevent double spaces where an invisible character sat next to an existing space. * Transforms do not appear in the AI Assistant chat – they run locally and show a brief confirmation dialogue. *** ## See Also * [QuickLauncher](/trados/quicklauncher/) * [Prompts](/trados/settings/prompts/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) # Troubleshooting Solutions to common issues with the Supervertaler for Trados plugin. *** ## Plugin not loading **Symptoms:** The TermLens panel does not appear in Trados Studio, or the plugin ribbon tab is missing. **Solutions:** 1. **Check Trados version** –the plugin requires **Trados Studio 2024** or later 2. **Verify .NET Framework** –ensure **.NET Framework 4.8** is installed on your system 3. **Reinstall the plugin** –remove the plugin via **Trados Plugin Management**, restart Trados, then install it again 4. **Check for errors** –open **Trados Plugin Management** and look for error messages next to the Supervertaler plugin entry Note After installing or updating the plugin, always restart Trados Studio completely (close all windows, not just the project). *** ## The installer finishes, but the plugin never appears **Symptoms:** You install from the App Store, the Trados Plugin Installer runs through every step and reports success, and then Supervertaler is not in the **View** menu and not in **Plugin Management**. Reinstalling does exactly the same thing. Often the plugin was working before and disappeared on its own. This is the most common installation problem, and it has one cause far more often than any other. ### First: was Trados Studio really closed? **The installer fails silently if Studio is running.** Not with an error – it completes normally and reports success, having changed nothing that matters. Each install location holds the plugin twice: the package, and an **unpacked** folder that Studio loads from and holds open the whole time it is running. An installer that cannot replace the unpacked folder leaves the old one in place, and **Studio will not re-extract into an unpacked folder that already exists**. The result is a successful install that changes nothing – even with the cleanup checkbox ticked, because that cleanup cannot delete files that are in use either. So: 1. Close Trados Studio – **every** window, not just your project. 2. Open Task Manager (`Ctrl+Shift+Esc`), look under **Details** for **`SDLTradosStudio.exe`**, and end it if it is still there. Studio can take a few seconds to exit, and can occasionally linger after its windows have closed. 3. Run the installer again, leaving **“Remove this plugin from all installation folders”** ticked. 4. Start Studio. That is the whole fix in most cases. ### Second: is there more than one copy in the Packages folder? Look in the folder for your Studio version (full paths further down): ```plaintext ...\Trados\Trados Studio\19\Plugins\Packages\ ``` There should be **exactly one** file with Supervertaler in the name. If there are two, that is the problem, and it explains why reinstalling never helps. Both files declare themselves as the same plugin, so Studio finds two copies of one plugin and loads neither. And because each install only ever replaces the file it is named after, a second install lands beside the first rather than on top of it – the fault repairs itself only if you delete **both** by hand and then install once. This can happen if the plugin has reached your machine by more than one route: from the App Store, from a file sent to you directly, or from an earlier version whose file was named differently. Delete every Supervertaler file you find there, then install once from the App Store. ### Third: is your Studio version greyed out in the installer? On the installer’s first screen, the list of installed Studio versions greys out any version the plugin does not support: > Studio versions that are not compatible with the plugin will be grayed out. **Almost always this means you downloaded the other build.** Supervertaler for Trados ships as two separate downloads, and the App Store page asks which you want: > Trados Studio - 2026 Release, 19.x Trados Studio 2024, 18.x The 2024 build greys out Studio 2026, and the 2026 build greys out Studio 2024 – by design, since each is built against its own Studio. If the version you want is greyed out, go back to the [App Store page](https://appstore.rws.com/plugin/432), click **Download**, and pick the other entry from the dropdown. If you are certain you downloaded the right build and your Studio version is *still* greyed out, email with the version number from **Help → About**. That means your Studio is outside the range this build declares, which is a fault at our end, not yours. ### Fourth: is it installed but switched off? Open **Help → Plugin Management** and look for Supervertaler in the list. If it is there but disabled, or shows an error beside it, that is a different problem from a failed install – the entry’s error message is the thing to send us. ### Last resort: clean out every copy by hand Only needed if the steps above do not work. Close Studio, then paste each of these paths into the address bar of a File Explorer window and delete anything with **Supervertaler** in the name – a `.sdlplugin` file in the `Packages` folders, a folder in the `Unpacked` ones. Replace `` with your own Windows user name. Use the block for your Studio version – `18` is Studio 2024, `19` is Studio 2026. Each path is one of the three choices the Trados Plugin Installer offers under “Please select the folder where the plugin will be installed”, so if you remember which you picked, start with that one – but check all three, because an earlier install may have used a different one. **Trados Studio 2024** ```plaintext C:\Users\\AppData\Roaming\Trados\Trados Studio\18\Plugins\ <- "All your domain computers" (the default) C:\Users\\AppData\Local\Trados\Trados Studio\18\Plugins\ <- "This computer for me only" C:\ProgramData\Trados\Trados Studio\18\Plugins\ <- "This computer for all users" ``` **Trados Studio 2026** ```plaintext C:\Users\\AppData\Roaming\Trados\Trados Studio\19\Plugins\ <- "All your domain computers" (the default) C:\Users\\AppData\Local\Trados\Trados Studio\19\Plugins\ <- "This computer for me only" C:\ProgramData\Trados\Trados Studio\19\Plugins\ <- "This computer for all users" ``` Inside each one there are two folders that matter, and **both** need clearing: * `Packages` holds the installed `.sdlplugin` file. * `Unpacked` holds the files Studio actually loads, extracted from it. This is the one people miss, and it is the one that blocks a reinstall: Studio will not re-extract over an `Unpacked` folder that is already there. Note **Two things that catch people out.** `AppData` is hidden by default, so you cannot browse to it – paste the whole path into the File Explorer address bar instead and press Enter. You may also see these written as `%AppData%` and `%LocalAppData%`. Those are just Windows shortcuts for `C:\Users\\AppData\Roaming` and `C:\Users\\AppData\Local`, and you can paste them into the address bar in place of the first part of the path if you prefer – they save typing your user name. Several of these will not exist, or will contain nothing from Supervertaler. That is normal – move on to the next. Delete **every** copy you find, including any that look like the current version: a leftover that looks correct is exactly the kind that wins the race against the new install. Then start Studio once and close it again, so it writes out a plugin list with nothing from Supervertaler in it, and install once more. Note Your settings, termbases, prompts, memory banks and licence key live in your Supervertaler data folder, not in these plugin folders. Deleting the plugin from all of them and reinstalling does not touch any of your work – see [Data Folder](/trados/data-folder/). Tip Still stuck? Email with your Studio version from **Help → About** and a screenshot of the installer’s first screen taken *before* you click Next. Those two things separate every remaining explanation. *** ## A keyboard shortcut does nothing **Symptoms:** A Supervertaler shortcut – e.g. `Alt+T` (translate segment), `Ctrl+Alt+T` (add term), `Ctrl+Alt+N` (non-translatable), `Ctrl+Alt+G` (AutoTagger), `Alt+Up` (quick-add to project termbase) or `Alt+Q` (QuickLauncher) – has no effect in the editor. **Solutions:** 1. **Clear the conflicting Trados default** – Trados Studio ships with its own actions bound to these key combinations, and the Trados binding wins. Go to **File → Options → Keyboard Shortcuts**, search for the conflicting Trados action, and delete its binding. The full table of what to delete is in [Keyboard Shortcuts](/trados/keyboard-shortcuts/#first-time-setup-free-up-trados-shortcuts) 2. **Repeat after a reinstall** – reinstalling or resetting Trados Studio restores its default bindings, so the shortcuts stop working again until you clear them once more *** ## ”Could not load SQLite” or DLL errors **Symptoms:** Error messages about missing DLLs or SQLite when opening settings or loading a termbase. **Solutions:** * **Restart Trados Studio** after the first install. The plugin pre-loads its own SQLite DLL to avoid conflicts with other plugins, but this requires a clean startup * If the error persists, reinstall the plugin to restore any missing DLL files *** ## Database locked / “cannot open database” **Symptoms:** Error when trying to load or write to the termbase database. **Solutions:** * **Close Supervertaler Workbench** if it has the same `.db` file open. Two applications writing to the same SQLite file simultaneously can cause lock conflicts * The plugin uses **read-only mode** where possible to minimise conflicts, but write operations (adding terms) require exclusive access * Verify the `.db` file is not on a drive that has gone offline (e.g., a disconnected network share) Caution If you share the database via a cloud-sync folder, ensure the file is fully synced before opening it in the plugin. Partially synced files can appear locked or corrupt. *** ## Terms not appearing **Symptoms:** TermLens shows no matches even though you know the segment contains terms that exist in your termbase. **Solutions:** 1. **Check the Read toggle** –open [TermLens Settings](/trados/settings/termlens/) and verify the termbase has **Read** enabled 2. **Verify the database path** –ensure the path points to the correct `.db` file 3. **Press F5** to force a full reload of your Supervertaler termbases from disk (note: F5 does not reload MultiTerm termbases) 4. **Reload the database** –click the **gear icon** in the TermLens panel to open settings, then close the dialogue. This forces a reload of the termbase data 5. **Check language pair** –the termbase source and target languages must match the current Trados project languages. Either direction works (the matcher handles inverted-direction termbases automatically), but the language pair itself must match. 6. **Check for reversed entries** –if a single specific term you know exists is silently not matching while other terms in the same segment do, the entry may be stored in the wrong direction in the database (e.g. Dutch text in the English column). This typically affects entries created or edited under v4.19.24 or earlier in projects whose direction was the inverse of the termbase’s. Open the **Termbase Editor** (double-click the termbase in TermLens Settings), find the term, check whether the source and target columns contain text in the expected languages, and use **Reverse source/target** to fix it. See [Termbase Management](/trados/termbase-management/) for details. *** ## MultiTerm terms not appearing **Symptoms:** Green chips from your MultiTerm termbases (`.sdltb` files) are not showing in TermLens, even though the termbases are attached to your Trados project. **Solutions:** 1. **Check your Trados project** –verify that MultiTerm termbases are attached via **Project Settings > Language Pairs > Termbases** 2. **Check the Read toggle** –open Supervertaler Settings (gear icon) and make sure the MultiTerm termbase’s Read checkbox is enabled 3. **Check languages** –the termbase’s source and target languages must match the current project’s language pair 4. **Navigate to another segment and back** to trigger a MultiTerm auto-refresh (F5 does not reload MultiTerm termbases – only segment navigation does) Note When you add terms in MultiTerm, navigate to a different segment in Trados to trigger the auto-refresh. TermLens checks for file changes on each segment change. See [MultiTerm Support](/trados/multiterm-support/) for full details. *** ## AI features not working **Symptoms:** Batch Translate produces no output, or single-segment translation returns an error. **Solutions:** 1. **Verify the API key** –open [AI Settings](/trados/settings/ai-settings/) and confirm the key is entered correctly with no extra spaces 2. **Check provider endpoint** –ensure the provider’s API endpoint is reachable from your network (no firewall or proxy blocking it) 3. **Ollama users** –make sure the Ollama service is running locally: ```bash ollama serve ``` Then verify the endpoint in AI Settings (default: `http://localhost:11434`) 4. **Custom provider** –double-check the endpoint URL and model name in the Custom OpenAI-compatible settings 5. **Check your API credits** –some providers return errors when your account balance is zero *** ## Database errors on Mac (Parallels) **Symptoms:** Database locked errors, “cannot open database”, or corrupt termbase data when running Trados Studio inside Parallels Desktop on a Mac. **Cause:** Your Supervertaler data folder is on a Mac-side shared path (e.g., `\\Mac\Home\Supervertaler`). Parallels mounts Mac folders as virtual network shares, and SQLite databases do not work reliably on network filesystems – WAL mode (used by Supervertaler termbases) requires a local filesystem for correct locking. **Solution:** 1. Move your data folder to the Windows side (e.g., `C:\Users\\Supervertaler`) 2. Copy your `.db` termbase files from the Mac-side location into the new Windows-side folder 3. Update the data folder path in Supervertaler settings, or delete `%AppData%\Supervertaler\config.json` and restart Trados to trigger the first-run setup again Note See [Installation – Running on a Mac (Parallels)](/trados/installation/#running-on-a-mac-parallels) for the recommended setup. *** ## Performance issues **Symptoms:** The editor feels sluggish, or TermLens takes a long time to display matches. **Solutions:** * **Large termbases** (50,000+ terms) may take a moment to index when the database is first loaded on startup. This is a one-time cost per session * **Close and reopen the editor** if the plugin feels unresponsive after a long session * **Disable unused termbases** –uncheck **Read** for termbases you do not need for the current project to reduce the matching workload * **Reduce batch size** in [AI Settings](/trados/settings/ai-settings/) if Batch Translate is slow or timing out *** ## Still having issues? 1. Ask a question in [GitHub Discussions](https://github.com/orgs/Supervertaler/discussions) – the community hub for both Supervertaler Workbench and Supervertaler for Trados 2. Check the [GitHub Issues](https://github.com/Supervertaler/Supervertaler-for-Trados/issues) for known bugs, feature requests, and workarounds 3. Open a new issue to report a bug or request a feature, including: * Your Trados Studio version * The Supervertaler plugin version * Steps to reproduce the problem * Any error messages or screenshots See [Support & Community](/trados/support/) for all the ways to get help. *** ## See Also * [Support & Community](/trados/support/) * [MultiTerm Support](/trados/multiterm-support/) * [TermLens Settings](/trados/settings/termlens/) * [AI Settings](/trados/settings/ai-settings/) * [Termbase Management](/trados/termbase-management/) * [Data Folder](/trados/data-folder/) # Token Usage & Costs Supervertaler keeps a **persistent log of the AI tokens and cost** of every operation, and a built-in **Usage & Costs report** to total and export it. This is the place to answer questions like *“how much did this project cost me?”*, *“how many tokens did we use this month?”*, or *“what should I bill this client for AI?”* – and it works for every provider, including custom and self-hosted models. It complements the **[Reports](/trados/reports/)** tab (which shows each call live, as you work) and the **[AI Cost Guide](/trados/ai-cost-guide/)** (which explains how costs work). The usage log is the durable, after-the-fact record. Note The usage log records **metadata only** – model, token counts, cost, project, file and language pair – and **never the prompt or response text**. That keeps the file small and safe to open in a spreadsheet or hand to an institution’s monitoring team. *Added in v4.20.56.* ### The usage log file Every AI call appends one line to a monthly file in your [Supervertaler data folder](/trados/data-folder/): ```plaintext …\Supervertaler\trados\usage\usage-2026-06.jsonl ``` The file is **JSONL** (one JSON object per line), so you can open it directly in Excel, parse it with a script, or load it into a notebook. A single line looks like this: ```json {"ts":"2026-06-18T16:07:03Z","product":"trados","task":"BatchTranslate", "provider":"claude","model":"claude-sonnet-4-6","project":"Example project (patent, en-nl)", "file":"US8312383.docx.sdlxliff","client":"","src_lang":"English (United States)", "tgt_lang":"Dutch (Belgium)","in_regular":654,"in_cache_read":0,"in_cache_write":27843, "out":808,"source":"actual","cost_usd":0.11849325,"cost_known":true,"duration_s":24.9,"ok":true} ``` A few things worth knowing: * **Every flow is covered** – Translate, Batch Translate, Quick Launcher, AutoPrompt, Proofread and Chat. A batch run is recorded as **one** line (the whole job), not one line per segment. * **`source`** is `actual` when the figures are the real token counts reported by the provider’s API, or `estimated` when they fall back to the chars/4 heuristic (see [Estimates vs actual cost](/trados/ai-cost-guide/#estimates-vs-actual-cost)). Cache reads/writes are broken out (`in_cache_read` / `in_cache_write`). * **`cost_known`** is `false` when the model isn’t in the price list – the **tokens are still logged**, the cost just shows as unknown until you add a rate (see [Custom and self-hosted models](#custom-and-self-hosted-models)). * It’s **on by default**. Turn it off any time in **Settings → AI Settings → “Keep a persistent token-usage log”**. ### The Usage & Costs report Open **Settings → AI Settings → “Usage & Costs report…”**. The window totals your logged usage and lets you slice it: * **Range** – This month, Last 3 months, This year, or All time. * **Group by** – **Project**, **Client**, **Model**, **Provider**, **Task** (Translate / Batch / Quick Launcher / …), **Day** or **Month**. Each row shows the number of calls, input and output tokens, total tokens, cost, and a **% actual** column – the share of that group backed by provider-reported figures rather than estimates. The footer shows the **range total** and your **month-to-date spend** (against your budget, if set). Note **Per-client billing.** Grouping by **Client** is only useful once each project has a client name. There is currently no in-app field for this – set it by adding a `"client": "Acme Ltd"` line to the project’s file under `…\Supervertaler\trados\projects\`. Projects without one are grouped under `(none)`. ### Exporting The report’s **Export CSV…** and **Export Excel…** buttons write the **detailed ledger** (one row per call, every column) for the selected range – ready for invoicing or analysis in any spreadsheet. CSV is written as UTF-8 (with a BOM, so Excel detects it correctly); Excel export produces a native `.xlsx`. Because both Supervertaler products write the **same columns**, an LSP can concatenate the CSV/JSONL from several translators – and from both Trados and Workbench – into one analysis. ### Monthly budget Set a soft monthly limit in **Settings → AI Settings → “Monthly budget (USD)”** (cents are allowed, e.g. `25.50`; `0` disables it). It is **advisory and never blocks**: once this month’s logged spend reaches the budget, starting a **Batch Translate** shows a *“Monthly budget reached – start anyway?”* prompt, so a large run can’t slip past unnoticed. Your month-to-date spend versus the budget is also shown in the Usage & Costs report. ### Custom and self-hosted models Costs are computed from a single price list, **`pricing.json`**, shared with Supervertaler Workbench. The bundled copy covers the built-in models. To price a **custom or self-hosted model** (or to override any rate for **both** Supervertaler products at once): 1. Copy the bundled `pricing.json` to `…\Supervertaler\pricing.json` (the shared data root), or create it there. 2. Add an entry under `models` keyed by the **exact model id** you use, with the input/output price per 1,000,000 tokens: ```json { "models": { "my-university-llama": { "input": 0.0, "output": 0.0 } } } ``` 3. Restart Supervertaler. Its cost now appears in the log and report; until then, its **tokens are still logged** with the cost marked unknown. Local models (Ollama) are priced at `0` – their token counts are still recorded, which is useful for capacity planning. ### See also * [AI Cost Guide](/trados/ai-cost-guide/) – how AI costs work, estimates vs. actual, provider dashboards * [Reports](/trados/reports/) – live, per-call token counts and cost as you work * [AI Settings](/trados/settings/ai-settings/) – where the toggle, budget and report button live * [Batch Translate](/trados/batch-translate/) – the main driver of token usage * [Data folder](/trados/data-folder/) – where the usage log and project files live # Voice Commands Control Trados Studio hands-free with spoken commands: confirm segments, navigate, insert TermLens matches, apply translation results, add terms and more – without touching your keyboard. Designed to pair with dictation tools such as Wispr Flow or Dragon: they type your translation, Supervertaler handles the commands. Note Selecting and replacing words you have just dictated – and handing the microphone to a dictation tool and back – has a page of its own: **[Dictation](/trados/voice-commands/dictation/)**. ### Starting and stopping Two ways to toggle voice commands: * Click the **🎤 microphone button** in the TermLens panel header (next to the ↻ refresh button) * Press **Ctrl+Alt+D** (also available in the editor right-click menu)  The 🎤 button in the TermLens header – green while listening The microphone button shows the state at a glance: | Colour | Meaning | | ---------- | -------------------------------------------------------- | | **Grey** | Off – click to start | | **Orange** | Starting (or downloading the voice runtime on first use) | | **Green** | Listening | Each command you speak flashes briefly in the TermLens status label (e.g. `🎤 "confirm"`), so you always know what was heard. Note **First activation** downloads the offline voice engine and a small English model (\~50 MB, one-time) – progress is shown in the status label. Every later activation is instant. If the TermLens panel isn’t open, a small floating status strip appears instead (bottom-right). You can drag it anywhere – the position is remembered. ### Default commands Everything works out of the box – no configuration needed. Most commands also respond to an alias, so you can use whichever phrasing comes naturally: | Say | Or | Action | | -------------------------- | ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | | ”select …" | "choose …” | Select those words in the target – see [Dictation](/trados/voice-commands/dictation/) | | ”delete that" | "delete this”, “remove that” | Delete whatever is selected in the target | | ”undo that" | "scratch that”, “undo” | Undo the last change (Ctrl+Z) | | “dictate" | "stop now” | Hand over to an external dictation tool and take it back (**off by default** – see [Dictation](/trados/voice-commands/dictation/)) | | “confirm" | "confirm segment” | Confirm segment and move to next unconfirmed | | ”next segment" | "go down” | Move to the next segment (without confirming) | | “previous segment" | "go up” | Move to the previous segment | | ”go to the top" | "go to top” | Jump to the first segment (Ctrl+Home) | | “go to the bottom" | "go to bottom” | Jump to the last segment (Ctrl+End) | | “copy source" | "copy from source” | Copy source to target | | ”clear target” | | Clear the target segment | | ”term one” … “term nine” | | Insert TermLens match 1–9 (with [capitalisation adaptation](/trados/termlens/#automatic-capitalisation)) | | “match one” … “match nine” | | Apply Translation Results match 1–9 (Ctrl+1–9) | | “term picker" | "pick term” | Open [TermPicker](/trados/termlens/termpicker/) | | ”term popup" | "show terms” | Open the [TermLens popup](/trados/termlens/termlens-popup/) | | ”add term" | "new term” | Quick-add the selection to your write termbases (Alt+Down) | | “add project term" | "project term” | Quick-add the selection to the project termbase (Alt+Up) | | “translate" | "translate segment” | AI-translate the active segment | | ”concordance" | "search memory” | Concordance search on the selection (F3) | | “zoom in" | "bigger font” | Increase the editor font size (see setup below) | | “zoom out" | "smaller font” | Decrease the editor font size (see setup below) | | “escape" | "close window” | Close the focused popup or dialog | | ”stop listening" | "voice off” | Turn voice commands off | ### One-time setup for “zoom in” / “zoom out” Trados Studio’s font-size actions ship **without a default keyboard shortcut**, so these two commands need a one-time binding: 1. Go to **File > Options > Keyboard Shortcuts > Editor** 2. Scroll down to the actions named simply **Increase** and **Decrease** (note: the Keyboard Shortcuts page has no search box – you have to scroll) 3. Set **Increase** to `Ctrl+Alt+PgUp` and **Decrease** to `Ctrl+Alt+PgDn` 4. Click **OK** From then on, “zoom in” and “zoom out” control the editor font size hands-free. (Make sure **Adapt font sizes** is enabled under File > Options > Editor > Font Adaptation.) ### Safety and privacy * **Fully offline** – recognition runs locally on your machine (Vosk engine); no audio is ever sent anywhere. * **Grammar-constrained** – the recogniser listens *only* for your command phrases, which is what makes commands fast and reliable. Normal speech and dictation are ignored. * **Foreground guard** – commands only execute while Trados Studio is the active window. Speaking in another app can’t trigger anything (“stop listening” is the one exception – it always works). ### Customising commands Right-click the 🎤 button (or click ⚙ on the floating strip) to open the **Voice command settings** dialog (also reachable via the **?** in its title bar and **F1** for this help page):  Voice command settings – every phrase, alias and action is editable * Enable/disable individual commands * Edit spoken phrases and add aliases * Add your own commands, mapped to either: * a **keystroke** chord sent to Studio (e.g. `ctrl+enter`, `alt+up`, `f3`) – any Studio or Supervertaler shortcut works * an **internal** plugin action: `insert_term_1`…`insert_term_9`, `term_picker`, `termlens_popup`, `navigate_next`, `navigate_previous`, `stop_listening`, `select_phrase`, `delete_selection`, `dictate_toggle` A phrase containing `{phrase}` – as in `select {phrase}` – takes the words you say after it as its argument, rather than matching exactly. Only `select_phrase` uses this. Note After saving, the recogniser updates immediately – no restart needed. #### Choosing words for a command The recogniser works from a closed vocabulary: your command phrases, plus the words of the segment you are in. It has no option to return nothing, so it always picks the best match from that list – which means **every word you add competes for that sound in every segment**, including against the words of your own translation. The number of commands barely matters: a long list of distinctive phrases costs nothing. What costs you is a single command built from a short, everyday word. * **Prefer distinctive words.** “insert tracked change” is free. A command called “to” or “for” is expensive, because it competes with those words wherever they appear in your text. * **Untick what you do not use.** Disabled commands are removed from the vocabulary entirely, so turning off the “term …” or “match …” numbers you never reach makes the rest more reliable. * **Watch for one word that will not select.** If [selecting](/trados/voice-commands/dictation/) a particular word keeps failing, the cause is usually a command phrase that sounds like it. Homophones are the trap, not spelling. The built-in “term eight” and “match eight” commands put *eight* into the vocabulary of every segment, and *eight* sounds exactly like the article *a* – so for a while, “select a further” could not be made to work at all. Supervertaler now resolves the number words that have common homophones (*a/eight*, *to/two*, *for/four*, *one/won*) back to the word your segment actually contains, but a custom command can reintroduce the problem with a different word. Commands are stored in `trados/settings/voice_commands.json` in your Supervertaler data folder, in the same format as Supervertaler Workbench’s voice commands – so you can exchange command sets between the two products. Default commands added in plugin updates are **merged into your saved set automatically** – your customisations are never touched. Because of this, if you want to get rid of a default command, **untick it rather than delete it** (a deleted phrase would come back if a later update re-ships it). **Restore defaults** replaces everything with the built-in set, discarding your customisations. ### Troubleshooting * **“Voice commands could not start”** – check that a microphone is available in Windows sound settings, and that the first-run download completed (an interrupted download can be retried by simply starting voice commands again). * **A command isn’t recognised** – speak the phrase on its own, at normal pace. If a phrase never triggers, give it a more distinctive alias in Voice command settings. * **Corrupt model** – delete the `trados/voice/models` folder in your Supervertaler data folder; the next activation re-downloads it. ### See Also * [Dictation](/trados/voice-commands/dictation/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) * [TermLens](/trados/termlens/) * [TermPicker](/trados/termlens/termpicker/) # Dictation Note This page builds on **[Voice Commands](/trados/voice-commands/)** – start there for turning voice on, the microphone button, the full command list and how to customise commands. Voice commands drive Trados Studio. A dictation tool writes your translation. This page is about the seam between the two – selecting words you have just dictated, replacing them, and handing the microphone back and forth without touching the keyboard. Everything here works with any dictation tool. [Wispr Flow](https://wisprflow.ai/) is the one the feature was built and tested against, so it is used for the worked example, but Dragon, Windows Voice Access and the rest follow the same three setup steps. ### Why two tools at all The two jobs need opposite things from a recogniser. **Commands** need a closed vocabulary. Supervertaler’s recogniser is given a short list of phrases and hears nothing else, which is what makes “confirm” fire instantly and reliably, and what stops your prose from triggering anything. It runs offline, on your machine. **Prose** needs the opposite: an open vocabulary, punctuation, capitalisation, and – for most of us – a second language. That is a different kind of engine, and dictation tools already do it well. So Supervertaler does not try to transcribe your translation. It selects, deletes, undoes, and gets out of the way. ### Editing what you just dictated Three commands, all on by default: | Say | Or | Action | | ----------------- | ---------------------------- | ------------------------------------------------------------------------- | | ”select …" | "choose …” | Select those words in the target segment | | ”select source …" | "source …” | Select those words in the **source** segment (off by default – see below) | | “delete that" | "delete this”, “remove that” | Delete whatever is selected | | ”undo that" | "scratch that”, “undo” | Undo the last change (Ctrl+Z) | “select” is followed by the words you want, not by a fixed phrase: **“select the duty cycle”**, **“select sealing ring”**. While a segment is open, every word of its target is added to the recogniser’s vocabulary, so it can hear the words that are actually in front of you. A few things worth knowing, because they explain what you will see: * **Dropped words are tolerated.** Unstressed function words often do not survive recognition – “comprises at most” comes back as “comprises most”. The selection widens to the segment’s own text, so you still get *comprises at most*. * **Whole words only.** Saying “select the” selects the standalone *the*, never the three letters inside *further*. * **Say it again for the next one.** Where a phrase occurs more than once, the first is selected and the status strip says **2 of 4 – say again for the next**. Repeating the same phrase steps to the next occurrence, and wraps round at the end. Naming more words works too. * **It only searches the target.** Source text, other segments and the termbase are not candidates. If a phrase cannot be selected, the status strip says why rather than doing nothing – `no "confirm" in this segment`, or `"the" is inside another word – say more words`. A selection in a place you did not name would be worse than none, because “delete that” would act on it. Note “delete that” removes the space before the deleted words as well, where that is unambiguous, so deleting a word from the middle of a sentence does not leave a double space behind. ### Selecting in the source `select source …` does the same thing in the **source** segment – useful for looking a term up, running a concordance, or adding a source/target pair to a termbase without touching the keyboard: **“select source cockpitdisplay”** → **“select cockpit display”** → **“add term”** It is **off by default**, because the source is usually in another language and needs a voice model for it. Tick **`select source {phrase}`** in Voice command settings; the first time you use it, the model for your project’s source language downloads in the background (\~40 MB, once) and the status strip tells you when it is ready. Note Source selection is **read-only**. “delete that” refuses a source selection, and so does starting dictation – deleting source text would damage the segment, its translation-memory match and the document’s alignment. Everything genuinely useful on a source selection (lookup, concordance, add a term) reads rather than writes. #### Long compounds In Dutch, German and other compounding languages, the word you want is often one the recogniser has never seen: languages like these build words on demand, so no dictionary contains them all. `beschermingsperiode` and `infraroodtouchscreens` are not in any Vosk Dutch model, large or small – this is a property of the language, not a limitation to be fixed with a bigger download. Supervertaler works around it by offering the recogniser the *parts* of every long word in the segment, and accepting any part as naming the whole word. So **“select source bescherming”** selects `beschermingsperiode`, and saying the compound naturally works too, because whichever part is heard is enough. A word with no recognisable part at all – a foreign loanword, typically – still cannot be selected. Name a neighbouring word instead. ### Handing over to a dictation tool The **“dictate”** command starts your dictation tool, stands aside while you talk, and stops it when you say “dictate” again. It ships **switched off**, because it drives a tool most installations do not have – switch it on in Voice command settings once you have done the setup below. While dictation is running, Supervertaler ignores every command except the way back out (and “stop listening”, which always works). This is the part a general-purpose macro tool cannot do: because it was *your command* that started the dictation, Supervertaler knows you are in it, and a translation containing the word “confirm” cannot fire a command into your document mid-sentence. #### Setup, with Wispr Flow as the example **1. Give the dictation tool a hands-free shortcut.** The “dictate” command is shipped expecting **Ctrl+Win+Space**, which is one of Wispr Flow’s own hands-free defaults, so there is usually nothing to change. Push-to-talk will not do – it needs a key held down, and the point here is not to touch the keyboard. **2. Teach it to write a marker for your stop phrase.** When you say “dictate” to stop, the dictation tool is still listening, so the word lands in your text. Rather than trying to guess and delete it, have the tool write a fixed marker instead: * In Wispr Flow, open **Dictionary** and add an entry mapping **dictate** to **ZZEND**. Supervertaler waits for `ZZEND` to appear and removes it, along with the space before it. A marker is used rather than an empty replacement for two reasons: most tools will not map a phrase to nothing, and a marker is far more reliable – spoken, your stop word arrives formatted, capitalised and punctuated (`. Dictate.`), and every one of those variants collapses into one fixed string you can search for. **3. Switch the command on.** Right-click the 🎤 button → **Voice command settings** → tick **dictate**. If your tool uses a different shortcut, or a different marker, edit the command’s action rather than any settings screen: it reads `dictate_toggle:ctrl+win+space:ZZEND` – trigger first, marker second. `dictate_toggle:middleclick` is accepted for a tool that listens for a middle click. ### Putting it together A correction, hands-free, start to finish: 1. **“select the duty cycle”** – the words highlight in the target 2. **“dictate”** – your dictation tool starts listening 3. *“the duty factor”* – it types over the selection 4. **“dictate”** – it stops, the marker is cleaned up 5. **“confirm”** – segment confirmed, on to the next If step 3 comes out wrong, **“undo that”** puts it back. ### Limitations * **The stop phrase must be distinct from your prose.** It is a word your dictation tool is listening for, so choosing a word you dictate often will bite you. * **A word buried in an earlier word, at the very end of a segment, cannot be selected.** There is nothing after it to tell the two apart. Say two words instead. * **Selection follows the active segment.** Move to another segment and the vocabulary changes with it. * **A word the recogniser has never seen cannot be selected**, and nor can a compound none of whose parts it knows. Name a word next to it instead. * **The status strip is the feedback channel.** Keep the TermLens panel open, or use the floating strip, so you can see what was heard and what happened. ### See Also * [Voice Commands](/trados/voice-commands/) * [Keyboard Shortcuts](/trados/keyboard-shortcuts/) * [TermLens](/trados/termlens/) # Supervertaler Workbench Welcome to the help center for 🖥️ **Supervertaler Workbench** – a free, open-source translation application built by translators, for translators.  **Supervertaler Workbench** integrates AI-powered translation with traditional CAT tool workflows. It runs on Windows, macOS, and Linux. * **Translate with AI** – GPT-4, Claude, Gemini, or local models via Ollama * **Work with CAT tools** – import/export files from memoQ, Trados, Phrase, CafeTran * **Translation Memory** – fuzzy matching, TMX import, concordance search * **Terminology** – termbases, automatic term highlighting, TermLens * **Companion tabs** – Chat (AI conversation), SuperLookup, Clipboard manager, Voice * **Voice** – always-on voice commands and push-to-talk dictation for any application * **SuperLookup** – system-wide translation lookup (TM, termbase, MT, web) * **QuickTrans Popup** – always-on-top popup with simultaneous translations from every enabled provider * **Quality Assurance** – spellcheck, tag validation, non-translatables | Requirement | Details | | ----------- | ------------------------------------------------------------------------------------------------------------ | | **OS** | Windows 10/11, macOS, Linux | | **Python** | 3.10 or higher | | **License** | Free and open source (MIT) | | **Source** | [github.com/Supervertaler/Supervertaler-Workbench](https://github.com/Supervertaler/Supervertaler-Workbench) | Start here: [Quick Start Guide](/workbench/get-started/quick-start/) *** ## Supervertaler for Trados Looking for the **Trados Studio plugin**? It has its own help center: [**Supervertaler for Trados Help →**](https://docs.supervertaler.com/trados/) Both tools share the same SQLite-based termbase format (`.db`) – termbases created in one work in the other. *** ## Getting Help * Browse this help center using the sidebar * Report issues on [GitHub](https://github.com/Supervertaler/Supervertaler-Workbench/issues) * Ask in [GitHub Discussions](https://github.com/orgs/Supervertaler/discussions) * Visit [supervertaler.com](https://supervertaler.com) # Autoprompt AutoPrompt uses an LLM to analyse your current project and generate a comprehensive, project-specific translation prompt. The generated prompt embeds the document’s domain, language pair, termbase, confirmed translations, detected source defects, terminology collisions, and preference cascades – ready to use as the **Custom Prompt** for AI translation. ## When to use it AutoPrompt is most useful when: * You are starting a new project and want a strong starting prompt instead of writing one from scratch. * You are translating in an unfamiliar domain and want the LLM to identify the EPO / IFRS / Terminologia Anatomica / domain conventions for you. * You want the prompt to reflect terminology decisions already made elsewhere in the project (confirmed segment translations, attached termbase) without having to retype them. If you already have a hand-tuned prompt that works for your client, you do not need AutoPrompt – just keep using it. ## Launching it 1. Open the **✨ AI** tab. 2. In the **Prompt Manager** sub-tab, find Section 2 (**Custom Prompt**) on the left side of the panel. The right column of that section is labelled “Generate one automatically” and contains a single **✨ AutoPrompt** button. Click it. 3. Supervertaler briefly reads a sample of the document with the AI to detect its context, then an **“AutoPrompt – confirm context”** dialog appears. It shows the detected domain (e.g. *Marketing – creative marketing copy, playful tone*) with: * a **Domain** dropdown you can leave as-is or correct, and * an optional **Context briefing** box where you can type a short note (e.g. “creative copy, playful tone, keep product names untranslated”). Click **Generate** to proceed (this is the normal case – just press Enter), or **Cancel** to abort. Anything you type in the briefing is treated as authoritative and overrides the detected domain where they conflict. 4. A **“Generating AutoPrompt”** progress dialog appears with an indeterminate busy bar. Reasoning-capable models (Opus, GPT-5, etc.) take 1–3 minutes; the rest of Supervertaler stays responsive while you wait. The dialog has a working Cancel button – cancelling stops the response being processed, though it can’t actually abort the HTTP request in-flight on the provider’s side. 5. When generation finishes, a **“Save AutoPrompt”** dialog opens. It shows a read-only preview of the generated content, plus a **Name** field (pre-filled with your project name) and a **Folder** dropdown (defaulting to **Translate** but listing every other top-level folder in your library – editable, so you can type a brand-new folder name and it’ll be created on save). 6. Click **Save** to write the prompt to the library and set it as the **Custom Prompt ⭐** for this project. Click **Cancel** to discard – the generated content stays in the chat log above so you can copy it out manually if you want it as text but not as a file. ## What gets analysed AutoPrompt gathers the following from your project and sends it to your configured AI provider: | Data | Purpose | | -------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- | | **Full source document** (up to 50,000 characters) | Context classification, terminology extraction, defect detection, cascade detection, project-context summary | | **All termbase entries** | Locked termbase embedded in the generated prompt | | **Confirmed segment translations** | Used as TM anchors – the highest-authority style and terminology reference, because they are decisions you have already made for this exact document | | **Translation Memory entries** (if attached) | Style anchors – the LLM matches the register and lexical choices of validated TM pairs | | **Language pair** | Embedded in the generated prompt | Note For a typical 30,000-word document, AutoPrompt costs roughly $0.20–$0.25 with a Sonnet-class model, or $1.00–$1.15 with an Opus-class model. The full document is sent so the LLM can actually read it and detect real patterns rather than guessing from metadata. ## Source-aware pre-generation passes Before the meta-prompt is sent to the LLM, Workbench runs several lightweight passes against the source content and injects the findings into the meta-prompt. The findings tell the LLM what to look for and give it concrete document-specific anchors instead of generic scaffolding. ### Context detection (AI-based) When you click AutoPrompt, Supervertaler sends a sample of the source to the AI and asks it to classify the document into one of: * **Patent** – claims, embodiments, prior art, figures, patent conventions * **Legal** – contracts, clauses, statutory references, notarial titles * **Medical** – clinical terms, dosages, anatomical terminology * **Technical** – specifications, software terms, standards * **Financial** – figures, IFRS / GAAP, regulatory language * **Marketing** – brand, audience, campaign language * **General** – fallback for mixed or unclassified content The AI reads the actual text rather than counting keywords, so classification is reliable across languages and doesn’t get fooled by superficial cues (for example a creative text that happens to mention a “Fig. 1” is no longer mistaken for a patent). The detected domain is shown in the **confirm-context** dialog before generation, where you can override it or add a briefing (see [Launching it](#launching-it) above). Note Earlier versions used an editable keyword list under **Settings → Domain Detection** to drive this. That panel has been removed: the AI classifier makes it unnecessary, and if the classifier ever gets it wrong you simply correct the domain (or add a one-line briefing) in the confirm-context dialog. ### Terminology-collision detection A built-in helper scans for known cross-term collisions in the source – groups of source-language terms whose natural English candidates would all map to the same target. Currently the helper covers Dutch mechanical / patent vocabulary (the highest-traffic source-language case): the `mantel` / `huls` / `mantelbuis` / `beschermhuls` cluster, the `pijp` / `buis` / `flexibele buis` cluster, the `voorzijde` / `voorvlak` / `achterzijde` distinction, and the `as` (axle vs geometrical axis) homograph. Each detected collision is presented with the EPO-conventional resolution so the LLM-generated prompt locks the correct mapping rather than picking arbitrarily. For any other source language or any other domain, the meta-prompt instructs the LLM to **perform its own collision scan** using patterns appropriate to the actual source language and detected domain – with explicit examples for medical (`arteria vs vena`, ligament vs tendon), legal (`agreement vs contract vs covenant`, liability vs responsibility), and financial (`revenue vs turnover vs sales`) collisions. The LLM-driven scan works for every language pair Workbench supports. ### Defect-detection pass A built-in helper extracts up to five verbatim defect examples from the source – hanging mid-sentence breaks ending in subordinating conjunctions, doubled spaces, plausible verb-ending typos (Dutch `-d` / `-t` confusion), and broken-compound double-space patterns. The Dutch-specific conjunction list catches the highest-traffic case; for other languages the meta-prompt instructs the LLM to perform the scan itself using equivalents in the actual source language (`weil` in German, `parce que` in French, `porque` in Spanish, `perché` in Italian, etc.). Quoting real defects in the generated prompt is far more effective than abstract “preserve defects faithfully” rules – the translator AI sees the actual surface forms it will encounter, not hypothetical examples. ### Preference-cascade extraction A built-in helper extracts up to three real `bij voorkeur ... bij nog meer voorkeur` cascades from Dutch sources (and `preferably ... more preferably ... even more preferably` from English). For other languages the meta-prompt instructs the LLM to look for source-language equivalents – `vorzugsweise / besonders bevorzugt` in German, `de préférence / plus préférablement` in French, `preferiblemente / más preferiblemente` in Spanish, `preferibilmente / più preferibilmente` in Italian, and `preferencialmente / mais preferencialmente` in Portuguese. Quoting one real cascade from the source anchors the generated prompt’s anti-truncation rule in a concrete example: “preserve THIS pattern, here is one from your own document”. ### TM-anchor wiring Confirmed source → target pairs from the project’s own segments are surfaced as TM anchors of the highest authority – they are locked decisions for this exact document. Pairs from any separately-attached `.tm` file are added with lower priority. If neither source has any pairs, the generated prompt’s “Previous Correct Translations” section is omitted entirely (rather than padded with “No TM data available”, which used to give the false impression that AutoPrompt had not even looked). ### Legal-entity scaffolding gate If the source contains no legal-entity markers (BV, NV, GmbH, Ltd., Meester, notaris, etc.), the generated prompt omits the BV/NV/Meester legal-entity-handling and statutory-reference sections. They are noise for a mechanical patent body where no entity names appear in running text, and they used to waste prompt-token budget that could have been spent on real document-specific anchors. ## Output format Generated prompts are formatted as proper Markdown – `##` headings for each major section, `-` bullet lists, `**bold**` for emphasised terms, a Markdown table for the project-specific termbase, and `---` horizontal rules between major sections. Open one in Obsidian, VS Code, GitHub, or any Markdown-aware viewer and it renders cleanly with a navigable outline. The Markdown markup is structural – it does not change what the translator AI does at translation time. The generated prompt’s inner OUTPUT FORMAT rule still says “translation only, no markdown formatting in the translation output”, so per-segment AI translations remain plain target text. ## Translator’s Comment methodology (always-on) Since v1.10.46, every AutoPrompt-generated prompt embeds the **Translator’s Comment** (TC) methodology by default, regardless of source language or domain. The methodology asks the translator AI to silently correct obvious mechanical defects in the source (typos, broken words, hanging mid-sentence breaks, doubled spaces, stray punctuation, reference-numeral mismatches that are unambiguous in context, missing diacritics, etc.) and append a single concise comment at the end of the segment in this exact format: ```plaintext ⟦TC: short factual description of the fix(es)⟧ ``` * The brackets are the mathematical white square brackets **U+27E6** (⟦) and **U+27E7** (⟧). These characters do not occur in source documents, so they are safe as out-of-band markers that can be extracted reliably in post-processing. * One marker per segment maximum; multiple fixes are joined with semicolons inside one marker. * Segments with no defects emit no marker. * When the translator AI inserts a word or short phrase to fill a clear gap, that supplied text is wrapped in standard ASCII square brackets `[like this]` inside the running translation, and the trailing marker references it (e.g. `⟦TC: [bracketed text] supplied to close hanging sentence⟧`). * Numerical values, dates, dosages, legal scope language, headings, identifiers, and proper names are never silently “corrected” – defects in those zones are preserved verbatim, with an optional `⟦TC: source ambiguous – ...⟧` marker if doubt exists. The defect categories that count as “obvious” are adapted to the actual source language by the LLM (Dutch -d/-t verb typos, German missing umlauts, French accent slips, Spanish/Italian conjugation typos, etc.). Note **The markers appear inline in the target text** – they are not yet auto-extracted into Workbench segment comments. Extraction into a dedicated comments pane is a separate follow-up. For now, you can copy or strip the markers manually, or run a downstream script that finds `⟦TC: ...⟧` regions and moves them into a structured comment field. Caution **Want a generated prompt without the TC methodology?** Edit the generated prompt after creation and remove the TRANSLATOR COMMENT FORMAT section plus any TRANSLATION MANDATE language about silent correction. A per-project opt-out via a UI toggle may be added in a future version – open an issue if you’d like to see it. ## Reviewing and refining the result The generated prompt appears in the prompt library tree and is loaded into the **Prompt Editor** automatically. You can: * **Read through the prompt** to verify it matches your project’s actual needs. * **Edit any section** directly in the editor – the generated prompt is just a regular `.md` file in your prompt library. * **Use AutoPrompt as a starting point** – the goal is to give you a high-quality first draft, not the final word. Adjust termbase entries, add client-specific quirks, tighten the register guidance. * **Re-run AutoPrompt later** if the project grows (more confirmed segments, more termbase entries) – each run is independent, so a later run produces a fresh prompt that reflects the project’s current state. ## Tips and limitations * **AutoPrompt cost scales with document size.** For very large projects (hundreds of thousands of words) the per-run cost can become significant. Consider running AutoPrompt against a representative subset rather than the full project for very large jobs. * **Termbase quality matters.** If you have an attached termbase that contains generic technical words (“system”, “board”, “installation”) that match almost any document, AutoPrompt will include those entries in the locked termbase and force the AI to follow them. Disable irrelevant termbases before running AutoPrompt. * **AutoPrompt does not guess.** If a pre-generation pass finds nothing in the source (no collisions for this domain / language combination, no defects, no cascades), the corresponding section is omitted from the generated prompt rather than padded with hypothetical examples. Hypothetical examples are worse than nothing. * **AutoPrompt and Supervertaler for Trados share the same prompt library folder.** A prompt generated in Workbench is immediately visible in the Trados plugin and vice versa. * **Vision-aware AutoPrompt is opt-in (since v1.10.178).** By default, AutoPrompt sends a text-only meta-prompt – figure *references* in the source are counted but the actual image files are not transmitted. Tick the **“🖼️ Include loaded figure images”** checkbox under the **✨ AutoPrompt** button in Section 2 to ship the figure images loaded in Section 4 alongside the meta-prompt. The LLM can then visually ground its terminology decisions – labelled parts, reference numerals, captions, visible components – directly into the generated terminology table. Three pre-flight gates apply: images must already be loaded in Section 4, the active model must support vision (Claude Sonnet/Opus 4.x, GPT-4o or newer, Gemini), and a confirmation dialog shows the estimated extra cost before the request goes out. Typical extra spend: roughly $0.05–$0.30 for 10–20 figures with a Sonnet-class model, or $0.25–$1.50 with Opus. For a high-value project (patent, technical spec), this is usually a worthwhile trade for a noticeably better-anchored generated prompt. ## See also * [Prompt Manager](/workbench/ai-translation/prompt-library/) – manage and organise generated prompts * [Creating Prompts](/workbench/ai-translation/prompts/) – write prompts from scratch * [AI Translation Overview](/workbench/ai-translation/overview/) – how the Custom Prompt is used during translation # AutoTagger AutoTagger looks at where the inline tags sit in the **source** segment and inserts that same set of tags into your **existing translation** at the right places – without changing any of the translated words. It is for the common case where a target has the correct translation but is **missing its tags or has them in the wrong spots** – typically after machine translation, pasting from another tool, or typing the target by hand. Those segments otherwise trip tag-validation checks even though the wording is fine. ## How it works 1. AutoTagger strips whatever tags are currently in the target. 2. It asks the AI to re-place the **full set of tags from the source** at the correct positions in your translation. 3. It **validates** the result before writing it: the tag set must match the source, the words must be unchanged, and the tags must be well-formed. 4. If the AI’s output doesn’t validate, it **retries once**. If it still fails, AutoTagger leaves the target **untouched** – so it never writes broken tags. Because the words are preserved exactly, AutoTagger is safe to run on a translation you have already reviewed. ## Running it on a single segment Run AutoTagger on the active segment one of three ways: * Click the **🏷️ AutoTagger** button on the editor toolbar. * Choose **Translate → 🏷️ Auto-tag Current Segment**. * Press **Ctrl+Alt+G**. Undo (`Ctrl+Z`) reverts it. ## Running it on many segments Use **Bulk Operations → 🏷️ Auto-tag Segments** to re-tag segments that are **already translated**. It runs over the selected segments (or all segments if none are selected) and the whole run is a single Undo. > **Note:** Ordinary AI translation (single-segment or Batch Translate) already places inline tags as part of the translation, so you do **not** need to run AutoTagger after translating. AutoTagger is for fixing tags on text that was translated some **other** way (MT, paste, hand-typed). The older “Fix tags with AutoTagger after translating” toggle in the Batch Translate dialog was removed in v1.10.319 for this reason. ## Configuring it Settings → **System Prompts** → **AutoTagger Instruction**. This editable template tells the AI how to place the tags. It supports these placeholders: | Placeholder | Meaning | | ----------------- | ---------------------------------------- | | `{{SOURCE_TEXT}}` | The source segment, with its inline tags | | `{{TARGET_TEXT}}` | Your current translation (tags stripped) | | `{{TAG_LIST}}` | The list of tags that must be placed | ## Tracking cost AutoTagger’s AI calls are logged under their own **“AutoTagger”** task in [Token Usage & Costs](/workbench/ai-translation/usage-costs/) (a bulk pass logs the tag-placement step as AutoTagger). Group the usage table by Task to see them broken out. *** ## See Also * [FuzzyFixer](/workbench/ai-translation/fuzzyfixer/) * [Tag Validation](/workbench/qa/tag-validation/) * [Batch Translation](/workbench/ai-translation/batch-translation/) * [Prompts](/workbench/ai-translation/prompts/) * [Token Usage & Costs](/workbench/ai-translation/usage-costs/) # Batch Translation Translate multiple segments at once with AI. ## Starting Batch Translation 1. **Select segments** to translate: * Click first segment, Shift+click last for a range * Or use **Edit → Select All** (`Ctrl+A`) 2. **Start batch**: * Press `Ctrl+Shift+T` * Or go to **Translate → Batch Translate** ## Batch Dialog Options ### Provider Selection Choose your LLM provider: * OpenAI (GPT-5.5, GPT-5.4 Mini) * Anthropic (Claude Sonnet 4.6, Claude Haiku 4.5, Claude Opus 4.8) * Google (Gemini 3.1 Flash-Lite, Gemini 2.5 Pro, Gemini 3.1 Pro) * Mistral, DeepSeek * Ollama (local models) ### Translation Mode | Mode | Description | | ---------------- | ------------------------------------------- | | **LLM Only** | Use AI for all segments | | **TM First** | Use TM matches above threshold, AI for rest | | **TM + Context** | Include TM matches as context for AI | ### Options * **Skip confirmed segments**: Don’t re-translate ✅ segments * **Include context**: Send surrounding segments for better quality * **Retry until complete**: Auto-retry segments that return empty ## Progress Tracking During translation: * Progress bar shows completion * Per-segment status updates * Can cancel anytime ## Retry Feature Enable **”🔄 Retry until all segments are translated”** to: * Automatically detect empty translations * Retry failed segments (up to 5 passes) * Ensure all segments get translated ## Tips ### Optimal Batch Size * 50-100 segments per batch works well * Very large batches may timeout * Split by page if needed ### Quality vs Speed * Claude Sonnet 4.6: Good all-round balance of speed and quality * GPT-5.5 / Claude Opus 4.8: Highest quality, slower and more expensive * GPT-5.4 Mini / Gemini 3.1 Flash-Lite: Fastest, lower cost ### Post-Edit Strategy After batch translation: 1. Review each segment 2. Fix any obvious errors 3. Confirm with `Ctrl+Enter` *** ## See Also * [AI Translation Overview](/workbench/ai-translation/overview/) * [Creating Prompts](/workbench/ai-translation/prompts/) * [Single Segment Translation](/workbench/ai-translation/single-segment/) # Chat The **💬 Chat** panel is a full AI assistant that supports OpenAI, Claude, Gemini, Ollama, and any OpenAI-compatible custom provider. It’s available in two places: as the **Chat** sub-tab of the **✦ AI** top tab, and as a **💬 Chat** tab in the editor’s right panel so you can keep a conversation visible while you translate. When you have **Supervertaler for Trados** running with its Assistant panel active, Chat automatically picks up the project context from Trados – active segment, surrounding segments, TM matches, termbase hits, project metadata – and answers questions about your real translation work without you having to switch out of Trados. This is especially useful on small laptop screens where there’s not enough room to keep the in-Trados Assistant panel visible alongside the editor. Summon Workbench, click the **💬 Chat** tab, and ask away while Trados stays in front for the actual editing. ## How it works Supervertaler for Trados runs a tiny localhost-only HTTP service called the **Supervertaler Bridge** while the AI Assistant panel is active. (The name is historical – it predates the Sidekick retirement in Workbench v1.10.4 and is kept stable because the Trados-side C# class looks up the bridge by that name.) Workbench’s Chat panel detects the bridge automatically and uses it to fetch the current Trados project state on every message you send. Nothing leaves your computer – the bridge listens only on `127.0.0.1`, requires a per-session authentication token, and is never reachable from outside the machine. ## The 🔗 Trados chip Above the chat input you’ll see a row of context chips (Document, TMs, Termbases, Files). When the Trados plugin is detected, a fifth chip appears: * **Hidden** until the bridge is detected for the first time. Users without the Trados plugin never see this chip. * **Lit green** – bridge is reachable; Trados context is included in chat messages. * **Greyed** – the bridge was previously available but is now unreachable (e.g. you closed Trados mid-session). The chip recovers automatically when Trados restarts. Click the chip to toggle it off if you want to ask a non-Trados-related question without the project context being included. The chip pref is shared across all chat views – toggling it in one place affects every send path. ## What the chat sees When the chip is on, every message you send to Chat is preceded by a context block that contains: | Field | What it is | | ------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Project name and file name** | So the AI knows what document you’re working on | | **Source and target language** | The language pair from Trados, used as a sanity-check against the chat’s reasoning | | **Active segment** | The source segment you’re currently editing in Trados, plus your current target draft (if any) | | **Surrounding segments** | A few segments before and after the active one – gives the AI enough context to know what kind of document this is (legal, technical, marketing, patent…) | | **TM matches** | Fuzzy and exact matches Trados has found for the active segment, with their match percentages and TM names | | **Termbase hits** | Term entries from your enabled termbases that match the active segment, including definitions, domains, and notes | This is the same context the in-Trados Supervertaler Assistant chat already uses, so answer quality is comparable. ## QuickLauncher prompts from Trados When **Supervertaler for Trados** is running, you can also send a Trados QuickLauncher prompt (Alt+Q from inside Trados Studio) over the bridge to be processed by Workbench’s Chat panel instead of the in-Trados Assistant. The Trados plugin builds a redacted *display* version (e.g. `[source document – N segments]` instead of the full project text) for the chat transcript and sends the fully-expanded prompt to the LLM. Workbench pops to the foreground on the Chat tab, the prompt and response render there, and the user can carry on the conversation as a normal Chat session. This is opt-in on the Trados side – a dropdown in Trados → **Settings → AI Settings → “QuickLauncher prompts go to:”** picks between the in-Trados Assistant (default) and Workbench Chat. Workbench is always ready to receive; nothing in Workbench needs to be configured. If Workbench isn’t running when Trados tries to send a prompt, Trados silently falls back to its own Assistant – the prompt is never lost. → [Supervertaler Bridge (Trados help)](https://docs.supervertaler.com/trados/ai-assistant/supervertaler-bridge/) – wire format and troubleshooting ## Example questions The Trados-aware chat shines when you ask questions that depend on what you’re actually translating right now: > *Which termbase entries apply to this segment?* The chat will list the termbase hits and explain how to apply them. > *What’s a more natural English translation for the segment I’m currently editing?* The chat will read the source, your draft target (if any), and any surrounding segments, and propose a polished translation. > *Are there any TM matches I should reuse for this segment?* The chat will summarise the TM matches and tell you which ones are close enough to confirm as-is. > *What domain is this document about?* The chat will infer the domain from the surrounding segments and reply. ## Privacy * The bridge listens **only on `127.0.0.1`** (loopback). Other devices on your network can never reach it. * Each Trados session gets a **fresh authentication token** – stale tokens from old sessions are useless. * The bridge **only starts when you have Assistant access** (paid subscription or trial). Without Assistant access, no bridge is started. * You can disable the bridge entirely on the Trados side by editing the plugin’s `settings.json` and setting `"sidekickBridgeEnabled": false`. See [Supervertaler Bridge](/trados/ai-assistant/supervertaler-bridge/) for details. ## Troubleshooting **The 🔗 Trados chip never appears.** The bridge isn’t being detected. Check: 1. Trados Studio is running with **Supervertaler for Trados v4.19.52 or later** 2. You have a paid subscription or active trial (the bridge is gated behind Assistant access) 3. You’ve **opened the Supervertaler Assistant panel** at least once in this Trados session – the bridge starts lazily when the panel is first activated 4. The handshake file `~/Supervertaler/trados/runtime/bridge.json` exists. If it doesn’t, check the bridge log at `%TEMP%\Supervertaler-bridge.log` for diagnostic information. **The chip appears but the answers don’t seem to use the project context.** Send a question that explicitly references your segment, like *“Which termbase entries apply to this segment?”*. If the answer cites specific terms from your termbases, the bridge is working. If the answer is generic, the chip may have been toggled off, or the bridge may have lost reachability – check the chip’s tooltip. **The chip is greyed out.** The bridge was reachable earlier but is no longer responding. Make sure Trados Studio is still running and the Supervertaler Assistant panel hasn’t been closed. The chip recovers automatically (within \~3 seconds) when Trados is reachable again. *** ## Related pages * [AI Translation Overview](/workbench/ai-translation/overview/) * [Supervertaler Bridge (Trados side)](/trados/ai-assistant/supervertaler-bridge/) * [Supervertaler (Trados)](/trados/ai-assistant/) # FuzzyFixer FuzzyFixer adapts an existing fuzzy TM match to the current source segment instead of translating it from scratch. When an 80–95% match is sitting right next to a segment, FuzzyFixer feeds that match into the AI prompt and asks the model to make only the edits the source change requires – keeping the existing wording, terminology, and tags wherever the source is unchanged. This is useful for repetitive, lightly-revised content (updated manuals, contract variants, new product revisions) where translating cold throws away a near-perfect human translation that is already in your TM. > Inspired by a suggestion from David Turnbull. ## Why use it By default the AI never sees a segment’s fuzzy TM match – it translates from the source alone, even when a high match is available. FuzzyFixer closes that gap: the model starts from the TM target and revises it minimally, so the result stays closer to your approved style and terminology than a fresh translation would. ## Running it on a single segment 1. Move to a segment that has a fuzzy match in range (see **Match range** below). 2. Run FuzzyFixer one of three ways: * Click the **🔧 FuzzyFixer** button on the Match Panel’s **TM Source** box (it is enabled only when the shown match is in range). * Choose **Translate → 🔧 Fuzzy Fix Current Segment**. * Press **Ctrl+Alt+F**. After it runs, the **TM Target** box shows a **track-changes view** of what the AI changed: the original TM target with the AI’s insertions underlined and removed words struck through, so you can see the edit at a glance before confirming. ## Using it in batch translation In the **Batch Translate** dialog, tick **🔧 Use FuzzyFixer**. When enabled: * The AI pass runs **per segment** (no batching), so each segment gets its own in-range fuzzy match injected into the prompt. * Segments **without** an in-range match are translated normally. * The toggle is remembered between sessions. Because it disables batching, FuzzyFixer batch mode is slower and costs more per segment than ordinary batch translation – use it on jobs that are genuinely revision-heavy. ## Configuring it ### Match range Settings → **AI Settings** → **🔧 FuzzyFixer**. FuzzyFixer only acts on matches inside a percentage range (default **75%–99%**). 100% exact matches are **never** altered. Lower the floor to catch looser matches, or raise it to limit FuzzyFixer to very close matches only. ### Instruction template Settings → **System Prompts** → **FuzzyFixer Instruction**. This editable template tells the model how aggressively to revise. In addition to the usual `{{SOURCE_TEXT}}`, it supports these placeholders: | Placeholder | Meaning | | ----------------- | ------------------------------------------------------------------ | | `{{TM_SOURCE}}` | The matched TM entry’s source text | | `{{TM_TARGET}}` | The matched TM entry’s target text (the translation being adapted) | | `{{MATCH_PCT}}` | The match percentage | | `{{TM_NAME}}` | The name of the TM the match came from | | `{{SOURCE_TEXT}}` | The current segment’s source text | ## Tracking cost FuzzyFixer’s AI calls are logged under their own **“FuzzyFixer”** task in [Token Usage & Costs](/workbench/ai-translation/usage-costs/). Group the usage table by Task to see them broken out from ordinary translation. *** ## See Also * [Fuzzy Matching](/workbench/translation-memory/fuzzy-matching/) * [AutoTagger](/workbench/ai-translation/autotagger/) * [Batch Translation](/workbench/ai-translation/batch-translation/) * [Prompts](/workbench/ai-translation/prompts/) * [Token Usage & Costs](/workbench/ai-translation/usage-costs/) # Image Context The **Image Context** viewer lets Supervertaler attach figure images to AI translation requests so the model can see the document’s drawings, photos, schematics or labelled parts in addition to the text. When a segment contains `Figure 1`, `Fig. 2A`, `Table 3`, etc., the matching image is shipped with the request so the AI’s translation is anchored in what the figure actually depicts – not just the words around it. The same viewer doubles as Supervertaler’s **DOCX image extractor**: point it at a Word document and it pulls every embedded image out into a numbered folder of PNG files, ready to load straight back as AI context. So the most common flow is one button-press: extract from a DOCX → context is auto-loaded → start translating. ## Where to find it The viewer is folded into the **Prompt Manager** (since v1.10.176; it used to be a standalone AI sub-tab): 1. Switch to the **✨ AI** tab → **Prompt Manager** sub-tab. 2. On the left of the panel, find **Section 4 – Image Context**. 3. Click the green **`Open ▸`** button. The right-hand panel swaps from the Prompt Editor to the Image Context viewer. 4. The viewer’s **`← Back to Prompt Editor`** button (or clicking any prompt in Section 5) returns you to the Prompt Editor. ## Extract images from a DOCX The viewer is a single toolbar at the top, with a results list + image preview below. 1. Add input files to the extraction queue: * **📄 Add DOCX** – pick one DOCX file * **📁 Add Folder** – add every DOCX file in a folder (batch input) 2. Choose an output strategy: * Enable **Auto-folder** to create an `Images` folder next to each DOCX * Or set a single output directory in the **Output directory** field 3. Set **Prefix** (default `Fig.`). 4. Click **🖼️ Extract Images**. 5. The freshly-extracted folder is automatically loaded as AI context for figure-aware translation – no second click required. 6. Use **📂 Extracted Files (click to preview)** below to preview images. ### Filename detection Since v1.10.190, extracted files are named after the **caption visible in the document** rather than sequential `Fig. N.png` numbers. So if your document labels figures as `FIG. 7`, the extracted file is `FIG. 7.png` – matching what the reader sees. The detector recognises the following label vocabularies (case-insensitive): | Pattern | Example filename | | -------------------------- | ----------------------------------- | | `FIG. N` / `FIGS. N` | `FIG. 7.png` (patent figures) | | `Figure N` / `Figs N` | `Figure 7.png` (academic / general) | | `Fig. N` | `Fig. 7.png` | | `Table N` | `Table 3.png` | | `Diagram N` | `Diagram 4.png` | | `Chart N` | `Chart 2.png` | | `Photo N` / `Photograph N` | `Photograph 12.png` | | `Scheme N` (chemistry) | `Scheme 1.png` | | `Plate N` / `Plate IV` | `Plate IV.png` (Roman numerals OK) | | `Exhibit A` (legal) | `Exhibit A.png` | The ID portion accepts `N`, `Na`, `NB` shapes (e.g. `FIG. 6a`, `Table 3B`). Documents that use Word’s built-in **Caption** paragraph style are detected even when no pattern matches – the leading sentence of the caption becomes the filename (capped at 80 chars). Images for which no caption can be detected fall back to the sequential `Fig. N.png` form. ### 🤖 AI label detection (opt-in) For documents that don’t follow the standard label vocabularies – marketing copy, blog posts, cookbooks, photo essays, foreign-language documents – tick **”🤖 AI label”** in the toolbar before clicking Extract Images. Each image that the text-pattern detector couldn’t label is sent to the active vision AI alongside surrounding text from the document, with a request to identify the figure’s label. **Requirements:** * A vision-capable model configured in Settings (Claude Sonnet/Opus 4.x, GPT-4o or newer, Gemini) * Internet connection **Cost:** roughly $0.005–$0.02 per AI-labelled image depending on provider – only spent on images the free text-pattern detector missed. A confirmation dialog showing the estimated extra cost appears before any API call is made; click Cancel to abort or run text-pattern only. **Pre-flight gates** that may interrupt the run: * No chat backend / API keys configured → friendly message, AI step skipped * Active model doesn’t support vision → choose between proceeding text-only or cancelling * Zero images need AI (all already pattern-detected) → silently skipped, no dialog, no cost If the AI replies `unlabelled` for a given image, that image falls back to the sequential `Fig. N.png` form. So the AI step is purely additive – it can only improve filenames, never make them worse. Note Documents with conventional captions (patents, academic papers, technical specifications) don’t need this feature – the free text-pattern detector handles them perfectly. The AI fallback is for documents where captions follow no recognisable pattern. ## Load a pre-existing folder of images If you already have a folder of images ready (e.g. one you extracted in an earlier session, or one you assembled by hand): 1. Click **📁 Load Folder** (the green button, to the right of Extract Images). 2. Pick the folder. 3. The images load into the AI context **and** populate the preview list below – just like a fresh extraction. Note **“Add Folder” vs “Load Folder”** – these are the two confusing buttons. **Add Folder** queues a folder of *Word documents* to extract images from. **Load Folder** loads a folder that already contains *image files* directly for AI context. The on-button tooltips spell out the distinction. ## Filename → figure-reference matching Supervertaler infers the figure reference from each image’s filename. Recognised patterns: | Filename example | Matched reference | | ---------------- | ----------------- | | `Figure 1.png` | `1` | | `Fig. 2A.jpg` | `2a` | | `figure3-b.png` | `3b` | | `Fig 10.tif` | `10` | When the AI later sees `Figure 1` or `Fig. 2A` in a segment’s source text, the matching image (case-insensitive, whitespace-/dash-/dot-normalised) is attached to the request. References that don’t match any loaded image are simply ignored – no error, no warning, the segment translates text-only. ## How loaded images reach the AI When AI translation runs (single-segment or batch), Supervertaler scans each segment’s source text for figure references. If a match is found AND a corresponding image file is loaded AND the active model supports vision (Claude Sonnet/Opus 4.x, GPT-4o or newer, Gemini), the image is base64-encoded (or passed as PIL data for Gemini) and attached to the request. Segments without figure references are translated text-only. A line appears in the log for every match, e.g. ```plaintext 🖼️ Detected figure references in segment #42: 1, 2a ✅ Including 2 figure images: 1, 2a ``` If the model is text-only (older GPT, Ollama models without vision, etc.), the figures are silently skipped and a warning goes to the log so you know why visual grounding didn’t fire. ## Using images with AutoPrompt (opt-in) Since v1.10.178, the loaded figures can ALSO be sent to the **AutoPrompt** generator – not just to the per-segment translator. Section 2 of the Prompt Manager has a companion checkbox under the **✨ AutoPrompt** button labelled **“🖼️ Include loaded figure images”**. Tick it before clicking AutoPrompt to ship the loaded figures alongside the meta-prompt. The LLM then uses the drawings to lock terminology decisions directly into the generated translation prompt – *“part 7 in Figure 1 is labelled ‘cylindrical sleeve’ → lock ‘mantelbuis’ → ‘cylindrical sleeve’ in the termbase”* – instead of having to guess from textual references alone. * **Off by default** – opting in is a deliberate per-project choice. * **Adds a small extra cost** – roughly $0.05–$0.30 for 10–20 figures with a Sonnet-class model, more with Opus. Negligible vs. the value of a typical project (a €1000 patent will spend less than 0.1% on visual grounding). * **Cost-confirmation dialog** pops up before the request so you can back out. See [AutoPrompt](/workbench/ai-translation/autoprompt/) for the full flow and pre-flight gates (vision-model check, missing-figures friendly message, etc.). ## Project persistence The currently-loaded folder path is saved into the `.svproj` file, so reopening a project automatically re-loads its image context. If the folder has moved or been deleted, you get a warning in the log and the project opens with no images loaded – re-pick the folder via **Load Folder** to restore. ## See also * [Prompt Manager](/workbench/ai-translation/prompt-library/) – where Section 4 (Image Context) lives * [AutoPrompt](/workbench/ai-translation/autoprompt/) – opt-in to ship figures with the meta-prompt * [AI Translation Overview](/workbench/ai-translation/overview/) – how images flow into per-segment translations # Using Local LLMs (Ollama) Ollama lets you run LLMs locally for privacy and offline translation. ## Install Ollama 1. Download and install from 2. Start Ollama 3. Pull a model (example): ```bash ollama pull llama3 ``` ## Use in Supervertaler Once Ollama is installed and running, Supervertaler can use it as a provider. Note Local models vary a lot in quality. For best results, test a few models on your typical content. # Overview Supervertaler integrates with leading AI language models for high-quality translation. ## Supported Providers | Provider | Models | | -------------- | -------------------------------------------------------------------------------- | | **OpenAI** | GPT-5.5, GPT-5.4 Mini | | **Anthropic** | Claude Sonnet 4.6, Claude Haiku 4.5, Claude Opus 4.8 | | **Google** | Gemini 3.1 Flash-Lite, Gemini 2.5 Pro, Gemini 3.1 Pro (Preview), Gemma 4 26B MoE | | **Mistral** | Mistral Large, Mistral Small | | **DeepSeek** | DeepSeek V4 Pro, DeepSeek V4 Flash | | **OpenRouter** | 200+ models via a single API key | | **Ollama** | TranslateGemma, Qwen 3, Aya Expanse (local, free) | | **Custom** | Any OpenAI-compatible endpoint | See [Supported LLM Providers](/workbench/ai-translation/providers/) for setup instructions for each provider. ## Quick Start 1. [Set up API keys](/workbench/get-started/api-keys/) 2. Open a project with segments to translate 3. Select a segment 4. Press `Ctrl+T` to translate ## Translation Methods ### Single Segment Translate one segment at a time: * Select a segment * Press `Ctrl+T` or click **Translate** button * AI translation appears in the target cell * Review, edit, and confirm ### Batch Translation Translate multiple segments at once: * Select segments (Shift+click for range) * Press `Ctrl+Shift+T` or use **Translate → Batch Translate** * Configure options in the dialog * All selected segments are translated Note Supervertaler has multiple batch scopes (selected / not-started / etc.). Start with **Translate → Batch Translate → Translate all not-started & pre-translated**. ### TM + AI Hybrid Combine Translation Memory with AI: 1. TM matches are checked first 2. High matches (e.g., >90%) are used directly 3. Lower matches are AI-translated with TM context 4. No matches use pure AI translation ## Prompts Prompts control how the AI translates. A good prompt includes: * Translation direction (source → target language) * Domain/subject matter * Style guidelines * Terminology rules * Special instructions ### Example Prompt ```plaintext You are a professional Dutch-to-English translator specializing in technical documentation. Maintain formal register. Use American English spelling. Keep all formatting tags like {1}, , in place. Translate naturally while preserving the original meaning. ``` See [Creating Prompts](/workbench/ai-translation/prompts/) and [Prompt Manager](/workbench/ai-translation/prompt-library/) for more. ## Provider Selection ### In Settings 1. Go to **Settings** tab 2. Find **LLM Settings** 3. Choose your preferred **Provider** and **Model** 4. Save settings ### Per-Translation When batch translating, you can choose the provider in the dialog. ## Quality Tips ### Get Better Results 1. **Use specific prompts** - Include domain, style, and rules 2. **Provide context** - Enable “include context” for surrounding segments 3. **Add termbase terms** - Attach terminology for consistent translations (see [Sending Terms to the AI](/workbench/termbases/ai-injection/)) 4. **Post-edit** - AI is great but not perfect; always review ### Common Issues | Issue | Solution | | ------------------ | ---------------------------------------- | | Wrong terminology | Add terms to termbase, include in prompt | | Inconsistent style | Be more specific in your prompt | | Tags removed/moved | Explicitly tell AI to preserve tags | | Too literal | Ask for “natural, fluent” translation | ## Cost Management ### API costs Cloud providers typically charge by usage (tokens). Pricing and free tiers change over time, so treat each provider dashboard as the source of truth. ### Reducing Costs 1. **Use Ollama** (local) when appropriate 2. **Translate only what you need** (for example not-started segments) 3. **Pre-translate with TM** when you have good matches 4. **Use smaller/faster models** for drafts, larger models for final passes *** ## Learn More | | | | --------------------- | ---------------------------------------------- | | **Single Segment** | [Translate one at a time →](single-segment.md) | | **Batch Translation** | [Translate in bulk →](batch-translation.md) | | **Creating Prompts** | [Write effective prompts →](prompts.md) | | **Local AI (Ollama)** | [Free, private AI →](ollama.md) | # Prompt Manager Supervertaler includes a Prompt Manager so you can create, organise, and reuse prompts across projects. It lives in the **✨ AI** tab → **Prompt Manager** sub-tab.  ## Layout The left side of the Prompt Manager is organised into **five numbered sections**, each with a coloured title strip. Read top to bottom, they are the four context layers that go into every AI request, followed by the library you pick prompts from: | # | Section | What lives there | | - | -------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | 1 | **System Prompt** | The built-in instructions for the AI. Auto-selected based on the current mode (Single Segment, Batch DOCX, Batch Bilingual). Click **View System Prompt** to see what it looks like and to jump to **Settings → 📝 System Prompts** if you want to edit it. | | 2 | **Custom Prompt** | Your project-specific instructions, in two columns: the active prompt on the left (with **Load External…** and **Clear**), and the **✨ AutoPrompt** button on the right. Set one from the library below, load an out-of-library file, or have the AI generate one. | | 3 | **Attached Prompts** | Optional extras stacked on top of the Custom Prompt. Right-click any prompt in the library and choose **📎 Attach to Active** to add it here. **Clear All Attachments** removes them all. | | 4 | **Image Context** | Visual references for the AI. Click the green **`Open ▸`** button to swap the right-hand panel from the Prompt Editor to the Image Context viewer, where you can extract images from a DOCX or load a pre-existing folder of figure images. Once loaded, images are sent as binary data alongside your prompt when figure references (Fig. 1, Figure 2A, …) are detected in a segment. The viewer’s **`← Back to Prompt Editor`** button (or simply clicking any prompt in Section 5) returns you to the Prompt Editor. | | 5 | **Prompt Library** | All your saved prompts. The button row below the heading lets you create new entries (**+ New**, **📁 New Folder**), refresh from disk, and collapse / expand every folder. | At the very bottom of the panel sits a single **👁 Preview Combined** button that opens a window showing exactly what will be sent to the AI for the current segment – the System Prompt, your Custom Prompt, every Attached Prompt, plus the segment text itself, all assembled in order. ## Setting the Custom Prompt Three ways to populate Section 2: * **From the library** – right-click any prompt in Section 5 and choose **⭐ Set as Custom Prompt**, or double-click it. The prompt name shows up next to the ⭐ icon in Section 2. * **From an external file** – click **Load External…** in Section 2 and pick any `.md` or `.txt` file from anywhere on your computer. The file stays where it is on disk; Supervertaler just references it. * **Have the AI generate one** – click **✨ AutoPrompt** in Section 2. The AI analyses your current document (domain, tone, terminology, confirmed translations) and produces a tailored prompt. See [AutoPrompt](/workbench/ai-translation/autoprompt/) for details. Whichever way you pick, the choice is saved into the `.svproj` immediately, so it survives a restart. ## Common uses * Maintain different prompts per client * Maintain different prompts per domain * Switch between “draft” and “final” translation styles * Pair a domain-specific Custom Prompt with one or two client-specific Attached Prompts (e.g. a “patents” Custom Prompt plus a “Client X house style” attachment) ## Tips * **Start simple** and evolve prompts as you learn what works for your language pair. * **AutoPrompt is a good starting point** for new projects, especially in unfamiliar domains – use the auto-generated prompt as a draft and edit it from there. * **Preview Combined is honest** – it shows you the actual final prompt that will be sent. If something looks wrong, it’s because something *is* wrong. * **External prompts can be edited in place** – if you load a prompt from an external file, Supervertaler can show it in the editor on the right and save changes back to the same file. ## See also * [AutoPrompt](/workbench/ai-translation/autoprompt/) – auto-generate a tailored translation prompt from the current document * [Creating Prompts](/workbench/ai-translation/prompts/) – what makes a good translation prompt when writing one by hand * [AI Translation Overview](/workbench/ai-translation/overview/) – how the assembled prompt is used during translation # Creating Prompts Prompts control how the AI translates (tone, domain, rules, formatting). ## What makes a good translation prompt * Target audience and tone (formal/informal) * Domain constraints (legal, medical, technical) * Rules for numbers, punctuation, terminology * Instructions to preserve tags and placeholders Caution If your text contains formatting or CAT tool tags, instruct the model to preserve them exactly. ## Quick checklist * Specify the language direction (source → target) * Specify style and audience (formal/informal, US/UK spelling, etc.) * Tell the model what to do with **tags/placeholders** (keep, don’t reorder, don’t delete) * Tell the model what to do with terminology (use termbase terms when provided) ## Next * [Prompt Manager](/workbench/ai-translation/prompt-library/) # Providers Supervertaler supports multiple AI providers so you can choose what fits your workflow and budget. You only need one to get started. ## Cloud providers ### OpenAI Models: **GPT-5.5**, **GPT-5.4 Mini** Get an API key at [platform.openai.com/api-keys](https://platform.openai.com/api-keys). GPT-5.4 Mini is a good starting point – fast, affordable, and high quality for most translation tasks. GPT-5.5 is the flagship for the most demanding work. ### Anthropic (Claude) Models: **Claude Sonnet 4.6**, **Claude Haiku 4.5**, **Claude Opus 4.8** Get an API key at [console.anthropic.com](https://console.anthropic.com). Claude Sonnet 4.6 is the recommended default. Haiku 4.5 is the fastest and cheapest option; Opus 4.8 is the most capable. ### Google (Gemini) Models: **Gemini 3.1 Flash-Lite**, **Gemini 2.5 Pro**, **Gemini 3.1 Pro (Preview)**, **Gemma 4 26B MoE** Get an API key at [aistudio.google.com/apikey](https://aistudio.google.com/apikey). Gemini 3.1 Flash-Lite offers a generous free tier and is a strong choice for high-volume work. ### Mistral AI Models: **Mistral Large**, **Mistral Small** Get an API key at [console.mistral.ai](https://console.mistral.ai). Particularly strong on European languages. ### DeepSeek Models: **DeepSeek V4 Pro**, **DeepSeek V4 Flash** Get an API key at [platform.deepseek.com](https://platform.deepseek.com). DeepSeek offers competitive pricing and strong multilingual quality. V4 Flash is the fast, cost-effective option for high-volume work. ## Gateway providers ### OpenRouter [OpenRouter](https://openrouter.ai) is an API gateway that gives you access to 200+ models from OpenAI, Anthropic, Google, DeepSeek, Mistral, Meta, and many others – all through a **single API key**. The model dropdown includes a curated selection for translation, and you can also type any OpenRouter model ID directly. Browse all models at [openrouter.ai/models](https://openrouter.ai/models). Get an API key at [openrouter.ai/keys](https://openrouter.ai/keys). Note OpenRouter adds a 5.5% platform fee on top of the underlying provider’s price. For most translation jobs this adds only a few cents. ## Local providers ### Ollama Run models entirely on your own machine – no API key and no internet required. See [Ollama Setup](/workbench/ai-translation/ollama/) for download and configuration instructions. ## Custom (OpenAI-compatible) For any provider that exposes an OpenAI-compatible API (Azure OpenAI, together.ai, local inference servers, etc.), select **Custom (OpenAI-compatible)** and enter the endpoint URL, model name, and API key. *** ## Related pages * [Setting Up API Keys](/workbench/get-started/api-keys/) * [Ollama Setup](/workbench/ai-translation/ollama/) * [AI Translation Overview](/workbench/ai-translation/overview/) # Single Segment Translation Use single-segment translation when you want maximum control: translate one segment, review, then confirm. ## Typical workflow 1. Select a segment 2. Run **Translate Segment** (shortcut: `Ctrl+T`) 3. Review the result 4. Edit if needed 5. Confirm the segment You can also trigger this from the menu: **Translate → Translate Current Segment**. ## Tips * Single-segment mode is great for tricky sentences and high-stakes text. * Use [SuperLookup](/workbench/superlookup/overview/) for research before confirming. # Token Usage & Costs Supervertaler Workbench keeps a persistent log of the AI tokens and cost of every operation, plus a built-in **Usage & Costs** report to total and export it. Use it to answer *“how much did this project cost?”*, *“how many tokens did we use this month?”*, or *“what should I bill this client for AI?”* – across every provider, including local and custom models. The log uses the **same format as the Supervertaler for Trados plugin**, so if you use both, their logs merge into a single analysis. Note The log records **metadata only** – model, token counts, cost, project, file and language pair – and **never the prompt or response text**. That keeps the file small and safe to share. ### The usage log file Every AI call appends one line to a monthly file: ```plaintext …\Supervertaler\workbench\usage\usage-2026-06.jsonl ``` It is **JSONL** (one JSON object per line) – open it in Excel, or parse it with a script. A line looks like: ```json {"ts":"2026-06-18T16:07:03Z","product":"workbench","task":"BatchTranslate", "provider":"claude","model":"claude-sonnet-4-6","project":"My Patent Job", "file":"","src_lang":"English","tgt_lang":"Dutch","in_regular":654, "in_cache_read":0,"in_cache_write":27843,"out":808,"source":"actual", "cost_usd":0.11849325,"cost_known":true,"duration_s":24.9,"ok":true} ``` It is **on by default**. Turn it off in **Settings → AI Settings → AI Cost Monitoring**. ### The Usage & Costs report Open **Tools → 💰 Token Usage & Costs…**. The window totals your usage and lets you slice it: * **Range** – This month, Last 3 months, This year, or All time. * **Group by** – Project, Client, Model, Provider, Task, Day or Month. Each row shows calls, input/output tokens, cost, and a **% actual** column (the share backed by provider-reported figures rather than estimates). The footer shows the range total and your month-to-date spend against your budget. ### Exporting **Export CSV…** and **Export Excel…** write the detailed ledger (one row per call) for the selected range – ready for invoicing or analysis. ### Settings & budget **Settings → AI Settings → AI Cost Monitoring** has: * **Keep a persistent token-usage log** – the on/off switch. * **Monthly budget (USD)** – a soft monthly limit (cents allowed; `0` disables). Once this month’s logged spend reaches it, starting a batch translation shows a warn-and-continue prompt. It is advisory and **never blocks**. ### Pricing custom / self-hosted models Costs come from a single price list, `pricing.json`, shared with the Trados plugin. To price a custom or self-hosted model – or override any rate for both products at once – copy the bundled `modules/pricing.json` to `…\Supervertaler\pricing.json` and add an entry keyed by the exact model id: ```json { "models": { "my-university-llama": { "input": 0.0, "output": 0.0 } } } ``` Until a rate is set, a custom model’s **tokens are still logged**, with the cost marked unknown rather than guessed. Local models ([Ollama](/workbench/ai-translation/ollama/)) are priced at `0`. ### How accurate are the figures? Every record is flagged **`actual`** or **`estimated`** in its `source` field – and the difference matters. * **`actual`** (the usual case): the token counts are the **exact numbers the provider’s API reported** for that call – the same numbers it bills you against. These are as accurate as it gets. This covers OpenAI, Claude, Gemini, Mistral, DeepSeek and OpenRouter, including the cached-token breakdown. * **`estimated`**: a fallback used only when the provider returned no usage data – currently **local models (Ollama)** and the occasional unparseable response. The estimate is a simple **characters ÷ 4** heuristic. It’s reasonable for English but can be well off for other content: scripts such as Chinese, Japanese, Korean, Arabic and Cyrillic pack a different number of characters per token, so the estimate often *under*-counts them. Ollama is free, so for local models only the token *count* is approximate – there is no cost involved. For **cost**, an `actual` record’s figure is the exact token count × the per-model rate from `pricing.json`, with **cached tokens priced at each provider’s cache discount** (e.g. Claude cache reads at 10% of the input rate and cache writes at 125%, OpenAI cache reads at 50%, Gemini 2.5+/3 at 25%). This matches how the Supervertaler for Trados plugin computes cost, so the two products agree. The one thing that can make it differ slightly from your provider’s bill: * The price list can **lag** a provider’s recent rate change. The model id and the token counts are still exact – only the unit price might be a little behind. For the definitive bill, your **provider’s own usage dashboard** is authoritative. The ledger is built for tracking trends, attributing usage to projects and clients, and capacity planning – where the exact, provider-reported token counts are precisely what you want. ### See also * [Batch Translation](/workbench/ai-translation/batch-translation/) – the main driver of token usage * [Supported LLM Providers](/workbench/ai-translation/providers/) – which providers report usage * [Using Local LLMs (Ollama)](/workbench/ai-translation/ollama/) – free, locally-run models * [General Settings](/workbench/settings/general/) – where AI Cost Monitoring lives # CafeTran Workflow Supervertaler supports CafeTran bilingual table DOCX workflows. ## Export from CafeTran 1. Open your project in CafeTran 2. Go to **Project → Export → Bilingual Table** 3. Choose **DOCX** format ## Import to Supervertaler 1. In Supervertaler: **Project → Import → CafeTran → Bilingual Table (DOCX)…** and select the exported DOCX 2. Translate and review in the grid 3. Confirm segments when ready (`Ctrl+Enter`) ## Export back to CafeTran 1. In Supervertaler: **Project → Export → CafeTran → Bilingual Table - Translated (DOCX)…** 2. Save the file ## Reimport to CafeTran 1. In CafeTran: **Project → Import** and select the bilingual table 2. Choose merge options as needed ## Notes * Preserve pipe-style markers and any CAT tags. * Don’t change the bilingual table structure in Word. # memoQ Workflow This guide covers working with memoQ bilingual files in Supervertaler. ## Export from memoQ ### Bilingual DOCX (Recommended) 1. In memoQ, open your project 2. Go to **Documents** view 3. Right-click your document → **Export Bilingual…** 4. Choose **Bilingual DOC/RTF/DOCX** 5. Select **Table format** (two columns) 6. Export the file Note The table format with source and target columns works best with Supervertaler. ### XLIFF Export 1. In memoQ, go to **Documents** view 2. Right-click → **Export Bilingual…** 3. Choose **memoQ XLIFF bilingual** 4. Save the `.mqxliff` file ## Import to Supervertaler 1. Go to **Project → Import → memoQ → Bilingual Table (DOCX)…** * Or **Project → Import → memoQ → XLIFF (.mqxliff)…** 2. Select your exported file 3. The segments appear in the translation grid ### What Gets Imported * ✅ Source text * ✅ Target text (if any) * ✅ Inline formatting tags (`{1}`, `[2}`, etc.) * ✅ Segment status ### memoQ Tag Handling memoQ uses special tag formats: | Tag Style | Example | Purpose | | --------- | ----------------- | --------------- | | Curly | `{1}` | Inline tag | | Mixed | `[2}` or `{3]` | Start/end tags | | Named | `{MQ}`, `{tspan}` | Formatting tags | These tags are highlighted in dark red in the grid (matching memoQ’s color). ## Translate in Supervertaler 1. Navigate through segments 2. Use AI translation (`Ctrl+T`) or translate manually 3. Confirm each segment (`Ctrl+Enter`) 4. Save your project regularly (`Ctrl+S`) ### Tips for memoQ Projects * **Preserve tags**: Keep all `{1}`, `[2}` tags in your translation * **Use SuperLookup**: Press `Ctrl+K` for TM and termbase searches * **Batch translate**: Select multiple segments and press `Ctrl+Shift+T` ## Export from Supervertaler 1. Go to **Project → Export → memoQ → Bilingual Table - Translated (DOCX)…** 2. Choose a filename 3. The bilingual table is recreated with your translations ## Import Back to memoQ 1. In memoQ, go to **Documents** view 2. Right-click your original document 3. Select **Import/Update Translation…** 4. Choose **From bilingual DOC/RTF/DOCX file** 5. Select the file exported from Supervertaler 6. Click **Import** ### Verify the Import * Check that translations appear in memoQ * Confirm status shows as “Translated” or “Edited” * Run memoQ’s QA to check for issues ## Complete Workflow ```plaintext memoQ: Export Bilingual DOCX ↓ Supervertaler: Import memoQ Bilingual ↓ Supervertaler: Translate (AI + manual) ↓ Supervertaler: Export memoQ Bilingual ↓ memoQ: Import/Update Translation ↓ memoQ: QA + Delivery ``` ## Troubleshooting ### Tags appear as plain text Make sure you exported as **Bilingual DOCX** (not “Export without tags”). ### Formatting lost on re-import This can happen if: * Tags were deleted or modified during translation * The bilingual table structure was changed **Solution**: Always keep tags exactly as they appear. ### Status not updating in memoQ memoQ’s import may not change segment status. You can: * Use memoQ’s filtering to find imported segments * Manually confirm segments in memoQ if needed *** ## See Also * [CAT Tool Overview](/workbench/cat-tools/overview/) * [Voice (commands and dictation in memoQ)](/workbench/voice/overview/) # CAT Tool Integration Overview Supervertaler is designed to work alongside professional CAT (Computer-Assisted Translation) tools, not replace them. Use it as a **companion tool** for AI-powered translation within your existing workflow. ## Supported CAT Tools | CAT Tool | Import Format | Export Format | | ---------------------- | --------------------- | ---------------------- | | **memoQ** | Bilingual DOCX, XLIFF | Bilingual DOCX | | **Trados Studio** | SDLPPX packages | SDLRPX return packages | | **Phrase (Memsource)** | Bilingual DOCX | Bilingual DOCX | | **CafeTran Espresso** | Bilingual table DOCX | Bilingual table DOCX | ## Why Use Supervertaler with CAT Tools? ### AI Translation Power CAT tools have limited AI integration. Supervertaler lets you: * Use multiple LLM providers (GPT-4, Claude, Gemini) * Create custom translation prompts * Batch translate with context awareness ### Workflow Flexibility * Translate offline with Ollama * Work on files while others are locked in the CAT tool * Quick review and post-editing without heavy software ## Typical Workflow ```plaintext ┌─────────────────────────────────────────────────────────────┐ │ YOUR CAT TOOL │ │ (memoQ, Trados, Phrase, CafeTran) │ │ │ │ 1. Receive project from client │ │ 2. Set up TM, termbases in CAT tool │ │ 3. Export bilingual file or package │ └─────────────────────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────────────────────────┐ │ SUPERVERTALER │ │ │ │ 4. Import bilingual file │ │ 5. AI translate + post-edit │ │ 6. Use SuperLookup for research │ │ 7. Export bilingual file │ └─────────────────────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────────────────────────┐ │ YOUR CAT TOOL │ │ │ │ 8. Import translations back │ │ 9. Run QA checks │ │ 10. Deliver to client │ └─────────────────────────────────────────────────────────────┘ ``` ## Key Concepts ### Preserving Formatting Supervertaler preserves CAT tool formatting tags: * memoQ: `{1}`, `[2}`, `{MQ}` inline tags * Trados: `<1>`, `` numbered tags * Tags are highlighted in the grid for visibility ### Segment Status Segment statuses map between tools: * **Draft** → Trados *Draft* / memoQ *Edited* * **Confirmed** → Trados *Translated* ✓ / memoQ *Confirmed* * **Approved** → Trados *Sign-off Approved* / memoQ *Reviewer 2 confirmed* See [Segment Statuses](/workbench/editor/segment-statuses/) for the full reference. ### Round-Trip Compatibility Files exported from Supervertaler can be imported back into the CAT tool with: * All translations preserved * Status information maintained * Formatting intact ## Choosing the Right Workflow | Scenario | Recommended Workflow | | ---------------------------- | ----------------------------------------------------- | | **Full project in memoQ** | [memoQ Bilingual DOCX](/workbench/cat-tools/memoq/) | | **Trados Studio package** | [SDLPPX/SDLRPX](/workbench/cat-tools/trados/) | | **Phrase/Memsource project** | [Phrase Bilingual DOCX](/workbench/cat-tools/phrase/) | | **CafeTran external view** | [CafeTran DOCX](/workbench/cat-tools/cafetran/) | | **Standalone DOCX** | Direct import, no CAT tool needed | *** ## Tool-Specific Guides | | | | ----------------- | ---------------------------------------- | | **memoQ** | [memoQ workflow guide →](memoq.md) | | **Trados Studio** | [Trados workflow guide →](trados.md) | | **Phrase** | [Phrase workflow guide →](phrase.md) | | **CafeTran** | [CafeTran workflow guide →](cafetran.md) | # Phrase (Memsource) Workflow Supervertaler supports Phrase (Memsource) bilingual DOCX round-trips. ## Export from Phrase 1. In Phrase Editor, go to **Document → Export → Bilingual DOCX** 2. Save the file Note Supervertaler works best with two-column bilingual files (Source/Target). ## Import to Supervertaler 1. In Supervertaler: **Project → Import → Phrase (Memsource) → Bilingual (DOCX)…** and select the bilingual DOCX 2. Translate (AI or manual) and review in the grid 3. Confirm segments when ready (`Ctrl+Enter`) ## Export back to Phrase 1. In Supervertaler: **Project → Export → Phrase (Memsource) → Bilingual - Translated (DOCX)…** 2. Save the file ## Reimport to Phrase Import the bilingual DOCX back into Phrase to update the document. ## Notes * Keep any tags/placeholders exactly as-is. * Don’t change the table structure in Word. Caution Don’t merge or split segments if you plan to reimport. # Trados Studio Workflow Note **Using Supervertaler for Trados (plugin)?** For documentation on the Trados Studio plugin (TermLens, AI Assistant, Batch Translate), see [Supervertaler for Trados](https://docs.supervertaler.com/trados/). This page covers the SDLPPX round-trip workflow using Supervertaler as a **standalone application**. This guide covers working with Trados Studio packages in Supervertaler. Tip **Just want to consult a Trados TM?** If you don’t need to round-trip a project – you only want to query a Trados `.sdltm` file from Supervertaler while it stays in use in Trados – you don’t need an SDLPPX package at all. See [Attaching a Trados TM (.sdltm)](/workbench/translation-memory/trados-sdltm/). ## Recommended: SDLPPX Packages The best way to work with Trados Studio is using **project packages** (SDLPPX files). ### Export from Trados 1. In Trados Studio, open your project 2. Go to **Project** → **Package** → **Create Project Package** 3. Configure package options (include all files) 4. Save as `.sdlppx` file ### Import to Supervertaler 1. Go to **Project → Import → Trados Studio → Package (SDLPPX)…** 2. Select your `.sdlppx` file 3. Supervertaler extracts and loads all segments Tip **Tip**: Package paths are saved in your `.svproj` file, so you can re-export to the same package later. ## Translate in Supervertaler 1. Navigate through segments 2. Use AI translation (`Ctrl+T`) or translate manually 3. Confirm each segment (`Ctrl+Enter`) 4. Tags appear as `<1>`, `` - keep them in your translation ### Trados Tag Format Trados uses numbered XML-style tags: | Tag | Purpose | | ------ | -------------------- | | `<1>` | Opening tag | | `` | Closing tag | | `<2/>` | Standalone/empty tag | These are highlighted in the grid for visibility. ## Export as Return Package 1. Go to **Project → Export → Trados Studio → Return Package (SDLRPX)…** 2. The return package is created with your translations 3. Segment status is updated to “Translated” Note The SDLRPX is created from your original SDLPPX with translations inserted. ## Import Back to Trados 1. In Trados Studio, go to **Project** → **Open Package** 2. Select the `.sdlrpx` file from Supervertaler 3. Trados imports the return package 4. Your translations appear in the target segments *** ## Alternative: Bilingual Review DOCX ⚠️ **Use SDLPPX instead if possible.** The Bilingual Review format has limitations. ### The Problem Trados Bilingual Review DOCX is designed for **review only**, not translation: * Empty target segments are not exported * You cannot translate from scratch using this format ### Workaround (if you must use it) 1. **In Trados**: Copy all source to target first * Edit → Task → Copy source to target (batch task) 2. **Export**: Project → Export → Trados Studio → Bilingual Review - Translated (DOCX) 3. **In Word**: Delete all target text (cells remain, but empty) 4. **Import to Supervertaler**: Project → Import → Trados Studio → Bilingual Review (DOCX) 5. **Translate** and export 6. **Re-import to Trados**: Merge into project Caution This workaround is tedious. Use SDLPPX packages whenever your client provides them. *** ## Complete SDLPPX Workflow ```plaintext Trados: Create Project Package (.sdlppx) ↓ Supervertaler: Import Trados Package ↓ Supervertaler: Translate (AI + manual) ↓ Supervertaler: Export Return Package (.sdlrpx) ↓ Trados: Open Return Package ↓ Trados: QA + Delivery ``` ## Troubleshooting ### ”Cannot find SDLXLIFF files” The SDLPPX might be corrupted or use an unsupported format. * Try re-exporting from Trados Studio * Ensure all files are included in the package ### Status shows “Draft” instead of “Translated” Fixed in **v1.10.259**. Earlier versions exported every segment as **Draft** regardless of its status in the grid, so even a fully-confirmed project arrived unconfirmed in Trados. Now a confirmed project exports with each segment marked **Translated** (Approved/Proofread map to *ApprovedTranslation*). Update to v1.10.259 or later and re-export the return package. ### Tags not matching Ensure you keep all `<1>`, `` tags in exactly the same positions. ### Source files not found on re-export If you moved your project, use **Project → Export → 🔗 Relocate Source Folder** to point to the new location of the SDLPPX. *** ## See Also * [CAT Tool Overview](/workbench/cat-tools/overview/) * [Import/Export Formats](/workbench/import-export/formats/) # Clipboard Manager The Clipboard Manager in Supervertaler Workbench captures everything you copy and keeps a persistent history that survives application restarts. Click any item to paste it; trigger a snippet or text conversion to paste the transformed result back to whichever app you came from. | How | Shortcut | | ------------------------------------------- | ------------------------------------------------------- | | Open Clipboard Manager from any application | **Ctrl+Alt+C** (⌘⌥C on macOS) | | Open Clipboard Manager tab manually | Click **📋 Clipboard Manager** in the Workbench tab bar | When you summon the Clipboard Manager via **Ctrl+Alt+C** from another app (e.g. Trados), Workbench automatically sends Ctrl+C in the source app *before* opening the tab. So you don’t need a separate “copy first” keystroke – the current selection lands at the top of the clipboard history the moment the tab opens. Note The tab was renamed from ”📋 Clipboard” to ”📋 Clipboard Manager” in v1.10.47 to match what the widget actually does – it has been more than a clipboard history for several versions (Snippets, Text Conversions, QuickLauncher Prompts, plus the clipboard history columns).  *** ## Three columns The tab is split into three side-by-side panels: * **📝 Text (left)** – plain text, rich text, and any other text copied from any application * **🖼 Images (middle)** – raster images (screenshots, copied graphics, etc.) * **📑 Menu (right)** – a tree of actions to apply to whatever’s currently on the clipboard: Personal Snippets, Special Characters, Text Conversions, and your QuickLauncher Prompts Each column has its own count in its header (e.g. *Text (37)*, *Images (8)*). A draggable splitter lets you resize the three panels. The column header whose widget currently holds keyboard focus is **highlighted in blue with an underline**, so it’s always obvious which column the arrow keys are steering. *** ## How clips are captured The Clipboard Manager monitors the system clipboard in the background. Every time you copy something in any application – a word in Trados, a URL, a code snippet, a screenshot – it is added to the top of the relevant list automatically. Duplicate copies of identical content are deduplicated (the existing item moves to the top instead of a new entry appearing). **Capacity limits:** | Kind | Maximum items | | ------ | ------------- | | Text | 200 | | Images | 50 | When a list is full, the oldest item is removed to make room. *** ## Privacy: controlling what gets captured Because the Clipboard Manager captures *everything* you copy while Workbench is running, it will also capture the username and password you copy out of your password manager, licence keys, and anything else you would rather it did not keep. Since **v1.10.369** there are three controls for this, in **Settings → 📋 Clipboard**. They can be combined, and every one of them saves and takes effect immediately – there is no Save button and no restart. | Control | What it does | | --------------------------------------------- | ---------------------------------------------------------------------- | | **Capture clipboard history** | Master switch. Off = nothing is captured at all. | | **Forget clipboard entries after a set time** | Deletes entries older than the window you set (1 minute – 7 days). | | **Never capture from these applications** | Ignores copying while a named application has the focus. Windows only. | ### The master switch Unticking **Capture clipboard history** stops capture completely. The check happens *before* Workbench reads the clipboard, so the content is never read, never hashed, never displayed and never written to the database – it does not enter Workbench’s memory at all. While capture is off, the Clipboard tab shows a red **⏸ Capture off** badge in its header, so an empty history is never mistaken for a fault. Note Switching capture off does **not** delete what is already there – existing clips stay in the history and can still be pasted. Use **Clear all** (see [Deleting clips](#deleting-clips)) if you want them gone. ### Automatic deletion Tick **Forget clipboard entries after a set time** and choose a window. Entries older than that are **deleted from the database**, not merely hidden from the lists. The sweep runs about once a minute, and also once when Workbench starts – so if Workbench was closed for three hours with a one-hour window set, the expired clips are gone before the list is ever drawn, rather than appearing briefly and then vanishing. ### Excluding particular applications Add process names exactly as they appear in Task Manager – `keepass.exe`, `1password.exe`, `bitwarden.exe` and so on. While one of those applications has the focus, copying is ignored entirely. The **Add common password managers** button fills in a dozen widely used ones in a single click. Matching is case-insensitive and the `.exe` is optional, so `KeePass` and `keepass.exe` both work. Caution This control is **Windows only** – identifying which application currently has the focus requires the Windows API. On macOS and Linux the list is saved but has no effect; use the master switch or automatic deletion instead. The Settings page says so on those platforms. Note **Why isn’t the exclusion list filled in by default?** Because exclusions are silent by design. If you copied a URL out of KeePass and it simply never appeared in your history, that would look like a bug rather than a feature. So the list starts empty and the password managers are one click away, chosen deliberately by you. ### Defaults Capture is **on**, automatic deletion is **off**, and the exclusion list is **empty**. An installation that never visits this page behaves exactly as it did before v1.10.369. *** ## Pasting a clip Click any item in the Text or Images list to paste it. What happens: 1. The item is placed on the system clipboard. 2. Workbench is hidden to the system tray. 3. `Ctrl+V` is sent to whichever window was active before the Clipboard tab opened. After pasting, the item is marked as used and appears greyed out. This makes it easy to track which clips you have already inserted in a session. Note **Latest clip is highlighted on open.** Every time you switch to the Clipboard tab, the most recent text clip (top of the list) is selected automatically – press **Enter** to paste it without touching the mouse. If you’d rather paste an older clip, arrow up/down to it first. ## The Menu column The third column gives you actions to apply to whatever’s on the clipboard. Expand a category by clicking its arrow or pressing **Right** with the category focused. ### 🔄 Refresh button The Menu column header has a small **🔄 Refresh** button on the right. Click it after editing any snippet `.md` file under `/snippet_library/` (rename a snippet, change a snippet body, add a new snippet, delete one) or any QuickLauncher prompt `.md` file in the shared prompt library. Refresh rebuilds the entire Menu tree from disk in one click – before v1.10.47 there was no way to pick up external file edits short of restarting Workbench. Refresh reloads three sources: the unified prompt library (via `UnifiedPromptLibrary.load_all_prompts()`), the snippet library (re-scans `/snippet_library/` with a fresh `SnippetLibrary` instance), and the Text Conversions table (in-code, but rebuilt for symmetry). ### 📌 Personal Snippets Your own text snippets (e.g. phone numbers, email signatures, boilerplate paragraphs). Snippets are loaded from `.md` files inside your user-data folder – see [Personal Snippets](/trados/text-transforms/) for the file format. Activating a snippet (click or Enter) copies its body to the clipboard and pastes it into the source app via the same hide-and-paste flow used for clipboard clips. ### ✨ Special Characters Quick-insert symbols, arrows, primes, dashes, quotes, currency signs, legal symbols, mathematical operators, and bullet characters. Activate one to paste the character into the source app. ### 🔁 Text Conversions Transform whatever text is on the clipboard. The conversions are computed against the *current* clipboard contents – so the typical flow is: select text in another app → **Ctrl+Alt+C** to open the Clipboard Manager (current selection auto-copies) → navigate to a conversion → Enter to paste the converted text back over your selection. #### The shipped defaults Eleven conversions ship out of the box: **Uppercase**, **Lowercase**, **Title Case**, **Sentence case**, **Single curly quotes**, **Double curly quotes**, **Round brackets**, **Square brackets**, **Remove soft hyphens (U+00AD)**, **Double quotes → single quotes**, **Make `bold`**. #### Adding your own Since v1.10.48, every text conversion is a `.md` file under `/text_conversion_library/`. The folder structure on disk is organisational – move files between folders to re-organise; the parent folder name becomes the conversion’s category. Drop a new `.md` file in the right folder, click **🔄 Refresh** on the Menu column, and the new conversion appears. Each file declares one conversion via YAML frontmatter at the top, with an optional human-readable notes section below: ```yaml --- type: wrap label: Mark as translator's comment ⟦TC: …⟧ prefix: " ⟦TC: " suffix: "⟧" --- Optional notes here. Workbench ignores everything below the closing ---. ``` The four supported `type` values cover most needs without arbitrary-code execution: | `type` | What it does | Required fields | Optional fields | | --------------- | ----------------------------------------------- | ---------------------------------------------------------------------------------------- | ----------------------------------------------------------- | | `case` | Change case of the whole text | `mode` (one of `upper`, `lower`, `title`, `sentence`, `swap`, `camel`, `snake`, `kebab`) | – | | `wrap` | Glue a prefix and suffix around the text | `prefix`, `suffix` | – | | `regex_replace` | Find/replace, literal or regex | `find`, `replace` | `regex` (default `true`), `case_sensitive` (default `true`) | | `strip_chars` | Remove every occurrence of any listed character | `chars` | – | Common optional metadata: * `label` – the display label shown in the Menu. Defaults to the filename stem if omitted (useful when the label contains characters that can’t be in filenames, like `:` or `"`). * `category` – overrides the folder-derived category. Set to an empty string to surface the conversion at the top level. * `enabled` – defaults to `true`. Set to `false` to hide without deleting (useful for project-specific conversions you might want back later). #### Concrete examples A wrap conversion for HTML emphasis: ```yaml --- type: wrap label: HTML prefix: suffix: --- ``` A strip-chars conversion that removes several invisible characters in one go (uses YAML’s `\u` escape inside double quotes): ```yaml --- type: strip_chars label: Strip invisible spaces (NBSP + figure space + narrow NBSP) chars: "   " --- ``` A regex find/replace for em-dash-to-en-dash: ```yaml --- type: regex_replace label: "Em dash (—) → en dash (–)" find: "—" replace: "–" regex: false --- ``` A regex find/replace using capture groups for UK → US “-our” → “-or” endings: ```yaml --- type: regex_replace label: "UK → US: drop the 'u' from -our endings" find: "([Cc]olo|[Ff]avo|[Hh]ono|[Ll]abo|[Nn]eighbo|[Bb]ehavio|[Ff]lavo|[Oo]do|[Rr]umo)u(r)" replace: "\\1\\2" regex: true --- ``` (Note the doubled backslashes in `replace` – `\\1` in YAML is needed to produce `\1` in the actual regex replacement string.) #### Things that DON’T work (and why) The four `type` values are deliberately limited – no arbitrary Python execution from user-data files, no shell commands, no network calls. If you have a transformation that genuinely needs Python (multi-step pipelines that produce intermediate state, calls to an external library, etc.), open a GitHub issue describing the use case and we’ll consider adding a `python_file` type with appropriate safeguards. Broken conversions (invalid `type`, bad regex, missing required field) are silently skipped on load and logged to the Workbench log – the clipboard flow never breaks on a typo. Fix the file, click 🔄 Refresh, and the conversion comes back. ### 💬 QuickLauncher Prompts Your custom AI prompts from the Prompt Manager, grouped by folder. Activating a prompt copies its body to the clipboard. ## Deleting clips **Single item** – right-click any entry in the Text or Images list and choose **🗑 Delete**, or select it and press the **Delete** key. **All clips** – click **Clear all** in the top-right corner of the Clipboard tab, or right-click any entry and choose **Clear all**. This removes the entire history from both the Text and Images lists and cannot be undone. (The Menu column is unaffected – it’s not history.) *** ## Keyboard navigation | Key | Action | | ------------------------------------- | -------------------------------------------------------------------- | | **Up / Down** | Move through items in the focused column | | **Right** | Move focus rightwards (Text → Images → Menu) | | **Left** | Move focus leftwards (Menu → Images → Text) | | **Right** on a Menu category | Expand the category | | **Left** on an expanded Menu category | Collapse it | | **Enter** | Paste the selected item / activate the selected action | | **Delete** | Remove the selected clip from history (Text / Images lists only) | | **Esc** | Hide Workbench to the system tray (when focus isn’t in a text input) | *** ## Empty state When a column contains no clips, a centred placeholder message is shown: * Text column: *No text yet – copy any text to start* * Image column: *No images yet – copy any image to start* *** ## Persistence The full clip history is stored in your user data folder in a shared SQLite database. Items are available the next time you open Supervertaler Workbench. Because the history is written to disk and survives restarts, it is worth deciding what you want captured in the first place – see [Privacy: controlling what gets captured](#privacy-controlling-what-gets-captured). *** ## Related pages * [Voice Commands & Dictation](/workbench/voice/overview/) * [Chat (AI conversation panel)](/workbench/ai-translation/chat/) * [Keyboard Shortcuts](/workbench/settings/shortcuts/) # Comments The **💬 Comments** tab in Workbench’s right panel is where per-segment annotations live. There are two sub-tabs: * **📝 Segment** – comments you author while translating: context notes, queries for the client, reminders to yourself, anchored highlights on specific words. * **✅ Proofreading** – AI-generated proofreading feedback, listed across the whole project and colour-coded by the LLM that produced it (read-only text; delete individually or all at once). Both are stored on the segment itself and persist in the `.svproj` file. Segment comments are **exported to the final document as Word comments** (yellow comment bubbles in the right margin of the exported `.docx`), with the range highlight covering exactly the words you anchored the comment to. Proofreading comments are review-only and don’t export. ## Segment comments ### Two kinds: segment-level and range-anchored There are two kinds of segment comment: * **Segment-level (no anchor)** – a general note about the whole segment. In the exported `.docx` the Word comment anchors to the entire paragraph. * **Range-anchored** – attached to a specific word or phrase you selected. In the exported `.docx` the Word comment highlights exactly those characters, the same way Trados and memoQ comments do. Both kinds coexist and you can have many of each on the same segment. ### Adding a range-anchored comment (Ctrl+M) 1. Click into the source or target cell of a segment. 2. Select the text the comment is about (e.g. `schroef`, `aangebracht`, or a multi-word phrase). 3. Press **Ctrl+M**. 4. A dialog opens showing which segment + which field (source or target) you’re anchoring to, plus a snippet of your selection for confirmation. Type the comment, click **OK**. The anchored text immediately gets a soft amber background in the editor cell so you can see at a glance which words have comments attached. The new comment also appears in the all-comments list (see below). Tip **Ctrl+M** matches memoQ’s “Add comment” shortcut. (Trados Studio uses **Ctrl+Shift+N** for the same action.) You can also add a comment without the keyboard: **right-click in the source or target cell → 💬 Add comment**. ### Adding a segment-level (unanchored) comment Place the cursor in the source or target cell **without selecting any text**, then press **Ctrl+M** (or right-click → **💬 Add segment comment**). The comment is attached to the whole segment rather than to a specific range. Useful for a general note, or for adding another comment to a segment without disturbing an existing one. ### The all-comments list The Segment sub-tab shows **one entry per Comment** – not per segment. A segment with three comments shows three entries, in document order. Each entry has: * A clickable **Segment #N** header. For anchored comments, the header reads `Segment #N ⚓ source` or `Segment #N ⚓ target` to tell you what kind of anchor it has. * For anchored comments: a quoted snippet of the anchored text, so you can see what the comment is *about* without jumping to the segment. * The full comment body. * A small footer line with the author and timestamp, plus a `(right-click for edit/delete)` hint. Clicking the **Segment #N** header jumps the grid to that segment (cross-page-aware – switches pagination first if needed). Conversely, **selecting a segment in the grid scrolls the list to – and highlights – that segment’s comment(s)**, so the active segment’s notes are always in view without hunting for them. **Right-click** on a Segment header (or anywhere on the entry) opens a context menu with **✏️ Edit comment…** and **🗑️ Delete comment**. Edit opens a small dialog pre-populated with the existing text; saving with an empty text field deletes the comment. Delete prompts for confirmation. The list rebuilds itself in real time as you add, edit, or remove comments. Note Earlier versions had a separate “Comment on current segment” box at the bottom of the tab. It only handled a single, unanchored note and overwrote the whole comment list when edited, so it was removed in v1.10.142. All comments – anchored and segment-level – now live in the one list, and you add them with **Ctrl+M** or **right-click → 💬 Add comment** in the editor. ### Editor visual cue: amber background Anchored comment ranges show a soft amber background in the source or target cell editors. This coexists with the existing syntax highlighting (tag pink, spellcheck underlines, etc.) – the amber is applied as a background colour on the anchored characters only. If you edit the target text after creating an anchored comment, the amber highlight stays at its original character offsets. If your edit shifts the anchor’s intended target, the highlight may end up on slightly-different wording. The simplest workaround is to edit the text first, then create the anchored comment. ### Reaching a comment from the grid You don’t have to open the Comments tab first to find a comment – you can get to it straight from the segment in the grid: * **Hover** the amber-highlighted range in a source or target cell to see the comment as a tooltip. * **Right-click** the highlighted range and choose **💬 Open comment**. Workbench switches the right panel to the **💬 Comments → 📝 Segment** sub-tab and briefly flashes the matching entry – handy when the Match Panel (or another tab) was showing. Segment-level comments have no highlighted range to aim at, so they’re reachable two other ways: * **Right-click anywhere** in a commented segment’s source or target cell → **💬 Open comment(s)** (opens the segment’s first comment). * **Hover or right-click the Status cell** of a commented segment: the tooltip lists the segment’s comments, and right-clicking offers **💬 Open comment(s)**. The Status cell is also **colour-coded** so you can tell comment types apart at a glance: a segment with a **segment comment** shows an **amber** background, one with a **proofreading comment** shows **purple**, and a segment that has **both** shows a **split amber|purple** background. ### Comments in exported documents When you export your project back to a Word document (**Project → Export → Export Translated Document…**), every segment comment becomes a Word comment in the output `.docx`. Behaviour per comment kind: * **Range-anchored to target text**: the Word comment’s range highlight covers exactly the anchored characters. If the anchor boundaries cut mid-run (e.g. inside a bold word), Workbench splits the run cleanly and preserves the formatting on both halves. Visually identical to a Trados or memoQ comment. * **Range-anchored to source text**: the source text isn’t in the exported (target-only) DOCX, so Workbench falls back to anchoring the comment to the whole paragraph, with the source snippet prefixed in the comment body. Example: `[Re: "schroef" (source)] translated as 'screw' based on context`. The reviewer reading the file sees the comment with the relevant source quote inline. * **Segment-level (unanchored)**: anchors to the whole paragraph, like a paragraph-wide annotation. The comment author defaults to the **Translator Name** field in **Settings → User Identity** – set that if you want comments attributed to your real name rather than your system username. Initials are derived automatically (multi-word names take the first letter of each word, e.g. `Michael Beijer` → `MB`; single-word names take the first two characters uppercased, e.g. `mbeijer` → `MB`). After export, Workbench logs how many comments were attached: ```plaintext ✓ Attached 4 segment comment(s) as Word comments (2 range-anchored) ``` If any comments couldn’t be matched to a paragraph (rare – usually because the target text was heavily reformatted by the Okapi merge step), the log says so and the export still completes – the comment-attach step is purely additive. ### Comments and bilingual table exports For bilingual-table export formats (Supervertaler Bilingual Table, memoQ, CafeTran, etc.), segment comments are written to a dedicated **Notes** column rather than as Word comments. Anchored and segment-level comments are concatenated into a single string in that column; the anchor metadata is **not** carried over (these formats don’t have a native concept of in-cell anchoring). Re-importing the bilingual table later preserves the comments as a single segment-level comment per segment. If you need range-anchored comments to survive a round-trip, export as DOCX rather than as a bilingual table. ## Proofreading comments The **✅ Proofreading** sub-tab lists AI-generated review feedback across the **whole project** – mirroring the Segment sub-tab’s all-comments list, so the two tabs now behave the same way. Generate the feedback with **QA ▸ Proofreading ▸ Proofread Translation…** (proofreading moved into the new [QA menu](#the-qa-menu) in v1.10.327). Each entry is one **(segment, model)** result: * A clickable **Segment #N · model** header that jumps the grid to that segment (cross-page-aware). * The proofreader’s findings, shown verbatim. The text is **read-only** – you read it and decide whether to act on it. * A **🗑️ delete** button that removes just that one comment. **Each LLM engine gets its own colour**, so if you ran the project through more than one model (e.g. GPT *and* Claude), you can tell at a glance which model flagged what. Selecting a segment in the grid scrolls the list to – and highlights – that segment’s proofreading comment(s), exactly like the Segment sub-tab. To clear everything at once, use **QA ▸ Proofreading ▸ Delete All Proofreading Comments**. Deletion is safe: re-running **Proofread Translation** regenerates the comments. Proofreading comments are **not** exported to the final document. They’re a translator-side review tool. ### The QA menu AI proofreading lives under the top-level **QA** menu (**QA ▸ Proofreading**), which also hosts **Delete All Proofreading Comments**. See **[AI Proofreading](/workbench/qa/proofreading/)** for how to run a pass. QA is Workbench’s home for quality-assurance features – see also [Spellcheck](/workbench/qa/spellcheck/), [Tag Validation](/workbench/qa/tag-validation/) and [Non-Translatables](/workbench/qa/non-translatables/). ## Quick reference | Action | Shortcut / How | | ----------------------------------------------- | ------------------------------------------------------------------------------------------------------------------- | | Add a range-anchored comment to selected text | **Ctrl+M** with text selected, or right-click → **💬 Add comment** | | Add a segment-level comment | **Ctrl+M** with no selection (cursor only), or right-click → **💬 Add segment comment** | | Jump to a commented segment | Click its **Segment #N** header in the all-comments list | | Open a comment from the grid | Hover the highlight for a tooltip; right-click the highlight, the cell, or the Status cell → **💬 Open comment(s)** | | Edit a specific comment | Right-click its header → **✏️ Edit comment…** | | Delete a specific comment | Right-click its header → **🗑️ Delete comment** | | Generate proofreading comments | **QA ▸ Proofreading ▸ Proofread Translation…** | | Delete one proofreading comment | **🗑️** button on its entry in the **✅ Proofreading** list | | Delete all proofreading comments | **QA ▸ Proofreading ▸ Delete All Proofreading Comments** | | Configure the author name for exported comments | **Settings → User Identity → Translator Name** | ## Related * [Editing & Confirming](/workbench/editor/editing-confirming/) * [Keyboard Shortcuts (Workbench)](/workbench/editor/keyboard-shortcuts/) * [Segment Statuses](/workbench/editor/segment-statuses/) # Editing & Confirming Translate by editing the **Target** column in the grid. ## Editing * Double-click a Target cell and type your translation. * Use standard editing shortcuts (undo/redo, copy/paste, find/replace). ### Multi-line target text * Use **Shift+Enter** to insert a line break inside the cell. ## Confirming segments Confirming matters for many workflows, especially when exporting back to a CAT tool. Typical workflow: 1. Translate (manual or AI) 2. Review the target text 3. Confirm the segment ### Confirm shortcuts * **Ctrl+Enter**: confirm the current segment (or all selected segments) and move to the next unconfirmed segment. * **Ctrl+Shift+Enter**: confirm all selected segments. You can also confirm by setting the segment **Status** dropdown to a confirmed status. ## Splitting and merging segments You can re-segment a document the way Trados Studio and memoQ allow – right from the grid: * **Split a segment:** click in the **Source** cell at the exact spot you want to divide, then right-click → **✂ Split segment here**. The existing translation stays with the first part; the second part starts empty, ready to translate. * **Merge two segments:** right-click in the **Source** cell → **🔗 Merge with next segment**. The two sources (and targets) are joined with sensible spacing, and the merged segment takes the less-complete of the two statuses. Both actions are fully **undoable** with **Ctrl+Z** (and redoable with **Ctrl+Y**). **When it’s available:** * Merge is only offered when the next segment is in the **same paragraph, table cell, file, and text unit** – so you can’t accidentally fuse separate paragraphs. If it isn’t allowed, the menu item is greyed out with the reason. * Split needs the cursor to land *inside* the source text, and is disabled while tags are shown in **compact** form or with **outer wrapping tags hidden** (switch to full tag view first, so the split lands in the right place). Note Split and merge are available for documents Supervertaler segments itself – **DOCX, IDML, HTML, PPTX, XLSX (via Okapi), and TXT/Markdown**. They are intentionally **hidden for bilingual CAT files** (Trados sdlxliff, memoQ/Trados/Phrase/Déjà Vu bilingual tables, PO), because those files have fixed segment slots owned by the other tool – adding or removing segments would break the round-trip back to that tool. (Trados and memoQ work the same way: you re-segment in their editor, which owns the segmentation.) ## Related pages * [Segment Statuses](/workbench/editor/segment-statuses/) – full reference for workflow statuses, match origins, and how they map to Trados and memoQ * [The Translation Grid](/workbench/editor/translation-grid/) * [Find & Replace](/workbench/editor/find-replace/) * [Tag Validation](/workbench/qa/tag-validation/) # Filtering Segments Filtering helps you focus on the segments you need right now. ## Common uses * Show only segments that contain a specific term * Focus on segments that need review * Quickly find repeated strings and fix consistency ## Tip: filter on selection A fast workflow is to select a word/phrase in the grid and use **Filter on selection**. ### Shortcut * **Ctrl+Shift+F** toggles filtering: * If a filter is not active, it filters on the current selection. * If a filter is active, it clears the filter. Note Filtering is designed to be fast even in large projects. # Find & Replace Supervertaler’s Find & Replace feature helps you quickly find text and make consistent changes across your translation. ## Opening Find & Replace * Press `Ctrl+F` or `Ctrl+H` * Or go to **Edit → Find & Replace** ## Dialog layout The action buttons are grouped to make destructive actions visually distinct from non-destructive ones. Reading left to right: * **Find next | Find all | Highlight all | Clear highlights** – non-destructive: searching and highlighting only, no edits made. * A vertical separator marks the boundary. * **Replace this | Replace all** – destructive: these modify your translation. They appear with an **amber background** as a visual cue that clicking them changes the document. The Close button sits at the far right. ### Keyboard shortcuts inside the dialog | Key | Where | Action | | ---------- | -------------------- | ------------------------------------------------------------------------------- | | **Enter** | in the Find field | Triggers **Find next** | | **Enter** | in the Replace field | Triggers **Replace all** (the existing confirmation prompt still appears first) | | **Escape** | anywhere | Closes the dialog | The Enter-in-Replace shortcut is intentional: the Replace all confirmation dialog catches accidental presses, so pressing Enter never silently overwrites your translation. ## Basic Usage ### Finding Text 1. Type your search term in the **Find** field 2. Click **Find all** to see all matches, or press **Enter** for Find next 3. Matches are highlighted in yellow in the grid 4. The results counter shows how many matches were found ### Replacing Text 1. Type your search term in the **Find** field 2. Type your replacement in the **Replace** field 3. Click the amber **Replace all** button, or press **Enter** while the Replace field has focus 4. Confirm the replacement count when the dialog asks 5. A confirmation shows how many replacements were made ## Search Options ### Match Three mutually exclusive modes (radio buttons): | Mode | Description | | ------------------ | ---------------------------------------------------------- | | **Anything** | Matches the search term anywhere in the text (the default) | | **Whole words** | Matches the search term only as a complete word | | **Entire segment** | Matches only when the whole segment equals the search term | ### Case sensitive * ✅ **On**: “Hello” won’t match “hello” * ❌ **Off** (default): “Hello” matches “hello”, “HELLO”, etc. ### Auto-adjust case When replacing, adjusts the replacement to match the case pattern of each match – ALL CAPS → uppercased, all lower → lowercased, Title Case → title-cased. It has no effect when **Case sensitive** is on, and is ignored in **Regex** mode. ### Search in | Scope | Description | | -------------------- | ------------------------------------------- | | **Source** | Search the source column | | **Target** (default) | Search the translations | | Both | Tick both boxes to search source and target | ### Reset edited to Draft When a replacement changes a target segment that was **Confirmed**, **Proofread** or **Approved**, that segment is reset to **Draft**, so the segments Find & Replace touched are easy to spot and re-check afterwards. * ✅ **On** (default): edited finished segments drop back to Draft. * ❌ **Off**: segments keep their existing status. This applies to **Replace this**, **Replace all** and **F\&R Sets** batch runs. Only segments whose text actually changes are affected, and replacing in the source column never changes a status. The setting is remembered between sessions, and `Ctrl+Z` restores the original text and status together. ## Regular expressions Tick **Regex** to treat the Find field as a regular expression (Python `re` syntax). * **Backreferences in Replace:** capture groups in the pattern can be reused in the Replace field as `\1`, `\2`, … (or `\g` for named groups). For example, Find `"([^"]+)"` and Replace `«\1»` turns `"events"` into `«events»`. * **Case sensitive** still applies (off = the whole pattern matches case-insensitively). * While Regex is on, the **Match** modes and **Auto-adjust case** don’t apply and are greyed out. * **Invalid patterns are caught:** an unbalanced pattern (e.g. `(`) or a bad backreference (e.g. `\9` with no matching group) shows a clear error and changes nothing – it never crashes or partially replaces. Tip A few handy patterns: `\s+` (runs of whitespace), `{2,}` (two or more spaces), `\b(\w+)\s+\1\b` (doubled words like “the the”), `+$` (trailing spaces). ## History Dropdowns Both the Find and Replace fields remember your recent searches: * Click the dropdown arrow to see your last 20 entries * Start typing to filter the history * History persists between sessions ## F\&R Sets (Batch Operations) Save and reuse multiple find/replace operations as a set. ### Creating a Set 1. Expand the **📁 F\&R Sets** panel 2. Click **➕ New Set** 3. Give your set a name (e.g., “Client Style Guide”) 4. The set appears in the dropdown ### Adding Operations to a Set 1. Enter your Find and Replace terms 2. Set your options (Match mode, Case sensitive, Regex, Search in) – these are all saved with the operation 3. Click **➕ Add Current to Set** 4. The operation is saved to the active set ### Managing Operations In the F\&R Sets panel: * **✓ (Enabled) column** – tick to include an operation when you click **Run All**; untick to skip it. (Hover for a reminder.) * **Edit** – double-click an operation to load it back into the Find/Replace fields. * **🗑 Delete Operation** – removes the selected operation from the set. * **🗑 Delete Set** – removes the selected set entirely. The **Match** column shows each operation’s mode, or **Regex** when the operation is a regular expression. ### Running a Batch 1. Select your set 2. Click **▶ Run All** 3. Each enabled operation runs in turn; regex operations (shown as “Regex” in the Match column) run with backreferences 4. See how many replacements were made Caution **An empty “Replace with” deletes matches.** An operation with a blank Replace field replaces every match with nothing – i.e. it deletes the matched text. If a set contains any such operation, Run All lists them and defaults the confirmation button to **No**, so you don’t wipe text by accident. ### Importing & exporting sets Share sets with colleagues: * **📤 Export** – save the selected set as a `.svfr` file * **📥 Import** – load a shared `.svfr` file ## Use Cases ### Terminology Consistency Create a set for client-specific terms: * “colour” → “color” (US spelling) * “organisation” → “organization” * “programme” → “program” ### Style Guide Rules Enforce style guidelines: * Double spaces → Single space * “e.g.” → “for example” * Straight quotes → Curly quotes ### Post-Translation Cleanup Clean up common MT artifacts: * Remove unwanted spaces before punctuation * Fix capitalization issues * Normalize formatting ## Tips Tip **Pro Tip:** Use regex mode for complex patterns. For example, `\s+` matches any whitespace to clean up extra spaces. Note **Undo Support:** All replacements can be undone with `Ctrl+Z` (within the same session). # Keyboard Shortcuts Master these shortcuts to work faster in Supervertaler. The exact keys are configurable in **Settings → Keyboard Shortcuts**; the tables below list the defaults. ## Navigation | Shortcut | Action | | ----------------------------------- | ----------------------------------------------------------------------- | | `↑/↓` | Previous/next segment when cursor is at the first/last line of the cell | | `Ctrl+Up` | Previous segment (always) | | `Ctrl+Down` | Next segment (always) | | `Ctrl+G` | Go to segment number | | `Page Up` / `Page Down` | Previous / next page (when paginated) | | `Shift+Page Up` / `Shift+Page Down` | Extend selection up / down by a screenful | | `Ctrl+Home` | First segment | | `Ctrl+End` | Last segment | ## Editing | Shortcut | Action | | ------------------------------ | ---------------------------------------------------------------------------------------------------- | | `Ctrl+Enter` | Confirm current (or selected) segment(s) and go to the next | | `Ctrl+Shift+Enter` | Confirm selected segments | | `Ctrl+Z` | Undo | | `Ctrl+Y` | Redo | | `Ctrl+C` / `Ctrl+V` / `Ctrl+X` | Copy / paste / cut | | `Ctrl+A` | Select all (in cell) | | `Shift+Enter` | Insert line break inside a cell | | `Tab` | Cycle between the source and target cells | | `Ctrl+Tab` | Insert a literal tab character | | `Ctrl+,` | Insert next tag / wrap selection with a tag pair (when available) | | `Ctrl+Shift+S` | Copy source text to target | | `Ctrl+M` | Add a comment to the selected source/target text (or a segment-level comment if nothing is selected) | | `Alt+D` | Add the word at the cursor to the custom dictionary | ## Translation | Shortcut | Action | | -------------- | -------------------------------------------------------- | | `Ctrl+T` | Translate current segment with AI | | `Ctrl+Shift+T` | Translate multiple segments | | `Ctrl+Space` | Insert the currently selected match from the grid | | `Alt+1…9` | Insert TermLens term #1…#9 (double-tap for #11…#99) | | `Ctrl+Alt+Q` | QuickTrans instant-translation popup (works system-wide) | | `Ctrl+Q` | Open QuickLauncher (AI prompt actions) | ## Find & Replace | Shortcut | Action | | ------------------------- | -------------------------------------- | | `Ctrl+F` | Open Find & Replace dialog | | `Ctrl+H` | Open Find & Replace on the Replace tab | | `Enter` (in the Find box) | Find next match | ## Filtering | Shortcut | Action | | -------------- | -------------------------------- | | `Ctrl+Shift+F` | Filter on selected text (toggle) | ## Lookup | Shortcut | Action | | ------------ | ---------------------------------------------------------------- | | `Ctrl+K` | Open SuperLookup (concordance) for the selection | | `Ctrl+Alt+L` | SuperLookup as a system-wide hotkey (works from any application) | ## View | Shortcut | Action | | ------------------------------- | ---------------------------------------------------------------------------------------------- | | `Ctrl+Plus` | Increase grid font size | | `Ctrl+Minus` | Decrease grid font size | | `Ctrl+Shift+=` / `Ctrl+Shift+-` | Increase / decrease results-pane font size | | `Ctrl+Shift+H` | Toggle tag view | | `Ctrl+Alt+P` | Toggle the [Document Preview](/workbench/editor/preview/) panel (and back to the previous tab) | ## Resources & Tools | Shortcut | Action | | -------------- | ---------------------------------------- | | `Ctrl+Shift+M` | TM Manager (separate window) | | `F5` | Force-refresh matches (clear cache) | | `Ctrl+Alt+C` | Open the Clipboard manager (system-wide) | ## File Operations | Shortcut | Action | | -------- | -------------------- | | `Ctrl+S` | Save project | | `Ctrl+O` | Open project | | `Alt+F4` | Quit the application | ## Termbase / Glossary | Shortcut | Action | | ------------------------------ | ----------------------------------------------------------------- | | `Ctrl+Alt+T` | Add the selected term pair to a termbase (opens the entry dialog) | | `Alt+Up` (or `Ctrl+Shift+1`) | Quick-add the selected term pair to the project termbase | | `Alt+Down` (or `Ctrl+Shift+2`) | Quick-add the selected term pair to the background termbase | | `Ctrl+Alt+N` | Add the selection to Non-Translatables | ## Voice (if enabled) | Shortcut | Action | | ------------------ | ------------------------------------------------------- | | `Ctrl+Shift+Space` | Voice dictation / push-to-talk (default – configurable) | | `Ctrl+Alt+O` | Toggle Always-On listening | *** ## Customising shortcuts You can view and customise every shortcut in **Settings → Keyboard Shortcuts**, and export a printable cheatsheet (HTML) or the raw definitions (JSON) from there. Note Some shortcuts match memoQ and Trados conventions (like `Ctrl+Enter` to confirm) to help translators who switch between tools. ## Tips ### memoQ-style navigation The arrow keys work like memoQ: * Press `↓` at the **last line** of a cell to move to the next segment * Press `↑` at the **first line** of a cell to move to the previous segment * Cursor column position is preserved when moving between segments ### Quick filtering 1. Select text in any segment 2. Press `Ctrl+Shift+F` to filter 3. Only segments containing that text are shown 4. Press `Ctrl+Shift+F` again to clear the filter # Navigating Segments Supervertaler is designed for fast, memoQ-style navigation in the translation grid. ## Moving between segments * Use the arrow keys to move within a cell. * When your cursor is at the **top** or **bottom** line of a cell, **Up/Down** can jump to the previous/next segment. ### Always move up/down If you want to move between segments regardless of where the cursor is inside the cell: * **Ctrl+Up** → previous segment * **Ctrl+Down** → next segment ### Jump to a segment * **Ctrl+G** opens **Go to Segment**. * Type a segment number and press Enter. ### Jump to start/end * **Ctrl+Home** → first segment * **Ctrl+End** → last segment ### Tab between Source/Target * **Tab** cycles between the Source and Target cell on the same row. * **Ctrl+Tab** inserts a literal tab character inside the text. ## Pagination By default all segments are shown on one page. You can split the project into fixed-size pages with the **Per page** selector (very large projects open paginated automatically). * See: [Pagination](/workbench/editor/pagination/) ## Tips * Keep one hand on the keyboard for speed: navigate, edit, confirm, repeat. * If you work in a CAT tool daily, customize shortcuts to match your muscle memory. See also: [Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) # Pagination By default Workbench shows **all** segments on a single page. You can split a project into pages of a fixed size using the **Per page** selector above the grid (5, 10, 25, 50, 100, 200, 500, or All). ## The default: All New projects open with every segment shown. With recent performance work this is comfortable well into the thousands of segments. The one exception is very large projects: a document with **more than 2000 segments** opens paginated at **500 per page** so the initial layout stays snappy. You can switch it back to **All** (or any other size) at any time with the **Per page** selector – your choice sticks for the rest of the session. ## Why you might still page * Work through a long project in smaller review batches * Keep the grid light on a low-spec machine or an enormous file ## Navigation Use the pagination controls to move between pages. ### Shortcuts * **Page Up** → previous page * **Page Down** → next page Note Go to Segment is pagination-aware: jumping to a segment on another page will switch pages automatically. # Document Preview The **Document Preview** is a reading view of your translation as a finished document, shown in the right-hand panel under the **📄 Preview** tab. As you translate, it rebuilds the document the way it will actually read – so you can sanity-check flow, paragraphing and formatting without exporting. ## What it shows * **Real document structure.** Sentences are reassembled into their original **paragraphs** (rather than one block per segment), and headings and lists are laid out accordingly – so the preview follows the source document’s structure, not the segmentation. * **Live translation.** Each segment shows its **target** text as soon as you translate it (falling back to the source until then), so the preview fills in as you work. * **Status at a glance.** Confirmed segments read clean; unconfirmed ones carry a faint amber tint. * **The current segment** is highlighted in light blue over its exact text, and the preview follows along as you move through the grid. ## Click to navigate Click any sentence in the preview to jump the grid straight to that segment – a quick way to move around a long document by reading rather than scrolling. ## Pop out into its own window Click **⧉ Pop out** at the top of the Preview tab to open the preview in a separate, resizable window – ideal for a second monitor. The pop-out window stays fully live: it tracks your edits, follows the current segment, and click-to-navigate still works. Close it to return to just the docked panel. ## Toggle with a keyboard shortcut Press **`Ctrl+Alt+P`** from the grid to switch the right panel to the Document Preview, then press it again to jump straight back to whatever tab you had open before (usually the Match Panel). Note The shortcut is configurable under **Settings → Keyboard Shortcuts** (View → “Toggle Document Preview panel”). See also: [Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) # Segment Statuses Every segment in Supervertaler carries two independent pieces of information: 1. **Match origin** – how the translation got there (TM match, machine translation, manual typing) 2. **Workflow status** – where the segment is in the review cycle (draft, confirmed, approved) These are two separate dimensions. A segment can be a 100% TM match *and* confirmed – the match tells you where the translation came from, and the workflow status tells you whether a human has signed off on it. *** ## Workflow Statuses These track a segment’s progress through the translation and review cycle. | Status | Icon | Meaning | | --------------- | ---- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Not started** | ❌ | No translation yet. The target is empty. | | **Draft** | ✏️ | Has a translation, but it has not been confirmed yet. This includes segments filled by AI, TM pre-translation, or manual typing that the translator has not yet explicitly confirmed. | | **Confirmed** | ✔ | The translator has explicitly confirmed the translation (Ctrl+Enter). The segment is saved to TM at this point. | | **Proofread** | 🟪 | A reviewer has checked and approved the translation. | | **Approved** | ⭐ | Final sign-off. The translation has passed all review stages. | | **Rejected** | 🚫 | A reviewer has rejected the translation. It needs to be revised. | | **Locked** | 🔒 | The segment cannot be edited. Typically used for non-translatable content or segments that must not be changed. | ### How confirmation works When you press **Ctrl+Enter**, the current segment is marked as **Confirmed** and the cursor moves to the next unconfirmed segment. Only confirmed segments are saved to Translation Memory. If you go back and edit a confirmed segment, it automatically drops back to **Draft** until you re-confirm it. This prevents accidental TM entries from half-finished edits. Note You can also change a segment’s status manually via the right-click context menu, or by selecting segments and using the **Status** menu in the menu bar. *** ## Match Origins Match origins tell you how the translation was initially obtained, before anyone reviewed it. They appear as the segment status when the segment has been pre-translated but not yet confirmed. | Status | Icon | Meaning | | ------------------ | ---- | ------------------------------------------------------------------------------------------------------------------------------------ | | **PM (102%)** | PM | **Perfect Match.** The segment matches a TM entry with identical context (surrounding segments). The highest level of TM confidence. | | **CM (101%)** | CM | **Context Match.** A 100% TM match where the preceding segment context also matches. | | **TM 100%** | ✅ | **Exact Match.** The source text matches a TM entry exactly, without additional context confirmation. | | **TM Fuzzy** | 🔶 | **Fuzzy Match.** The source partially matches a TM entry (typically 75-99%). The translation will need editing. | | **MT** | 🤖 | **Machine Translation.** Generated by an MT engine or AI model. | | **Repetition** | 🔁 | **Internal Repetition.** The same source text appears elsewhere in the project and the translation was auto-propagated. | | **Pre-translated** | ⚡ | Generic pre-translation from an unspecified source. | Perfect Match and Context Match are shown as small coloured **PM** and **CM** text badges in the status column, mirroring the way Trados Studio labels them, rather than as emoji icons. ### What happens when you confirm a match When you confirm a segment, its status changes from its match origin (e.g. “CM 101%”) to **Confirmed**. The match percentage is preserved internally, but the status icon now reflects the workflow state rather than the match type. This is the same behavior as in Trados Studio and memoQ. *** ## Mapping to Other CAT Tools Supervertaler’s statuses map cleanly to Trados Studio and memoQ. When you import a file from either tool, both the match origin and the workflow status are preserved. When you export back, statuses are mapped to the correct values for each tool. ### Workflow statuses | Supervertaler | Trados Studio | memoQ | | ------------- | -------------------- | -------------------- | | Not started | *(no translation)* | Not started | | Draft | Draft | Edited | | Confirmed | Translated ✓ | Confirmed | | Proofread | Translation Approved | Reviewer 1 confirmed | | Approved | Sign-off Approved | Reviewer 2 confirmed | | Rejected | Translation Rejected | Rejected | | Locked | Locked | Locked | Caution **Trados naming quirk:** In Trados Studio, “Translated” (green checkmark) actually means *confirmed by the translator*. This is equivalent to Supervertaler’s **Confirmed**. Trados “Draft” is equivalent to Supervertaler’s **Draft**. Note **A note on localised (translated) UIs.** When Supervertaler’s interface is shown in another language, the status *names* you see in the status column are **display labels only** – the underlying status is an internal value, and import/export mapping to Trados Studio and memoQ is keyed on that internal value, **not** on the visible text. So a confirmed segment shown as “Bevestigd” in Dutch, “Confirmed” in English, or any other localisation still maps to Trados “Translated” / memoQ “Confirmed” on export. Localising the labels never changes the data or the round-trip. If you’re translating the interface, two tips for the status labels: * **Mirror what the user’s other CAT tools call these states in the same language.** Trados Studio and memoQ ship localised UIs; aligning your status terms with theirs means the user sees consistent vocabulary everywhere. Inventing new terms is what causes confusion, not localising itself. * **Leave the match-origin shorthand untranslated** – `PM`, `CM`, `TM 100%`, `TM Fuzzy`, `MT`, and the percentages are industry-universal across tools and languages (and the PM/CM badges deliberately mirror Trados). ### Match origins | Supervertaler | Trados Studio | memoQ | | ------------- | ---------------------------- | --------------------- | | PM (102%) | Perfect Match | 102% (double context) | | CM (101%) | Context Match | 101% (context match) | | TM 100% | Exact Match (100%) | 100% | | TM Fuzzy | Fuzzy Match | Fuzzy (75-99%) | | MT | Machine Translation (NMT/AT) | MT | | Repetition | Repetition (auto-propagated) | Repetition | *** ## Related pages * [Editing & Confirming](/workbench/editor/editing-confirming/) * [The Translation Grid](/workbench/editor/translation-grid/) * [CAT Tool Integration](/workbench/cat-tools/overview/) * [Trados Studio Workflow](/workbench/cat-tools/trados/) * [memoQ Workflow](/workbench/cat-tools/memoq/) # The Translation Grid The translation grid is where you spend most of your time: each row is a **segment** (usually a sentence or paragraph) with source and target text. ## Columns The grid has five columns: | Column | What it is | | ---------- | -------------------------------------------------- | | **#** | Segment number (row index) | | **Type** | Segment type (depends on the file format/importer) | | **Source** | Original text (typically read-only) | | **Target** | Your translation (editable) | | **Status** | Segment status (dropdown) | ## Editing behavior * The grid is optimized for speed, but edits are intentionally lightweight. * **Double-click** a cell to edit. * Use **Shift+Enter** for a line break inside a cell (multi-line target). ## Confirming & status * Use the **Status** dropdown to set the segment state. * Keyboard confirm is supported (see [Editing & Confirming](/workbench/editor/editing-confirming/)). Common statuses include: * Not started * Translated * Confirmed * Proofread * Approved Note If you plan to reimport into a CAT tool, do not merge/split content across segments. Segment boundaries must stay compatible. ## Visual cues * **Tags** (CAT tool placeholders and formatting markers) are highlighted to make them hard to miss. * **Spellcheck** (if enabled) underlines misspelled target words. * **Termbase matches** can be highlighted in the source. ## TermLens panel placement You can dock the TermLens panel directly above or below the grid from the **View** menu: * **Show TermLens above grid** –places the panel between the filter bar and the grid. * **Show TermLens below grid** –places it under the grid. The change applies immediately, no need to reopen the project. Clicking the option that is already active hides the panel again. ## Splitting and merging segments Right-click in a **Source** cell to **✂ Split segment here** (at the clicked position) or **🔗 Merge with next segment** – Trados/memoQ-style re-segmentation, fully undoable. See [Editing & Confirming](/workbench/editor/editing-confirming/#splitting-and-merging-segments) for details and when it’s available. ## See also * [Navigation](/workbench/editor/navigation/) * [Editing & Confirming](/workbench/editor/editing-confirming/) * [Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) * [Filtering](/workbench/editor/filtering/) # Api Keys To use AI translation you need an API key from at least one provider. Enter your key in **Settings → AI Settings**. ## Supported providers | Provider | Where to get a key | | ------------------------------------- | -------------------------------------------------------------------- | | **OpenAI** | [platform.openai.com/api-keys](https://platform.openai.com/api-keys) | | **Anthropic (Claude)** | [console.anthropic.com](https://console.anthropic.com) | | **Google (Gemini)** | [aistudio.google.com/apikey](https://aistudio.google.com/apikey) | | **Grok (xAI)** | [console.x.ai](https://console.x.ai) | | **Mistral AI** | [console.mistral.ai](https://console.mistral.ai) | | **DeepSeek** | [platform.deepseek.com](https://platform.deepseek.com) | | **OpenRouter** (200+ models, one key) | [openrouter.ai/keys](https://openrouter.ai/keys) | | **Ollama** | No key needed – runs locally | ## Entering a key 1. Open **Settings → AI Settings** 2. Select your provider from the **Provider** dropdown 3. Paste your API key into the **API Key** field 4. Click **Test Connection** to verify 5. Save settings Keys are stored locally and are only sent to the provider’s own API endpoint. ## Switching providers You can configure keys for multiple providers in the same settings panel. Switch between them without re-entering credentials – the key for each provider is remembered independently. ## Using Ollama (no key required) Ollama runs models entirely on your machine. No API key or internet connection is needed. See [Ollama Setup](/workbench/ai-translation/ollama/) for download and configuration instructions. ## Using OpenRouter (one key for everything) If you prefer not to manage multiple accounts, OpenRouter lets you access 200+ models from all major providers with a single API key. Create an account at [openrouter.ai](https://openrouter.ai) and paste your key into the **OpenRouter** provider slot. ## Troubleshooting | Problem | Solution | | --------------------- | -------------------------------------------------------------------- | | ”Invalid API key” | Double-check the key; ensure no leading or trailing spaces | | ”Rate limit exceeded” | Wait a moment, or upgrade your API plan | | ”Model not found” | Check the model name in settings; it may have been updated | | No response | Check your internet connection and that the provider’s service is up | *** ## Next steps * [Create your first project](/workbench/get-started/first-project/) * [Supported LLM Providers](/workbench/ai-translation/providers/) * [AI Translation Overview](/workbench/ai-translation/overview/) # Your First Translation Project Let’s walk through creating a complete translation project from start to finish. ## Creating a New Project ### Option 1: Import a Document 1. Go to **Project → Import → Import Document…** (`Ctrl+O`) 2. Select your Word document 3. Choose the source language (e.g., “English”) 4. Choose the target language (e.g., “Dutch”) 5. Click **Import** Your document is now segmented and ready for translation. ### Option 2: Import a Text File 1. Go to **Project → Import → Text / Markdown File (TXT, MD)…** 2. Select your `.txt` file 3. Each line becomes a separate segment ### Option 3: Multi-File Project 1. Go to **Project → Import → Folder (Multiple Files)…** 2. Select a folder containing DOCX or TXT files 3. Choose which files to include 4. All files are imported as one project ## Understanding the Interface After import, you’ll see the main window. Along the top are the workspace tabs (**Editor**, **TMs**, **Termbases**, **AI**, **SuperLookup**, **Clipboard Manager**, **Voice**, **Settings**). The **Editor** tab holds the translation grid: ```plaintext ┌──────────────────────────────────────────────────────────────────────┐ │ Editor │ TMs │ Termbases │ AI │ SuperLookup │ Clipboard │ Voice │ ⚙ │ ├──────────────────────────────────────────────────────────────────────┤ │ # │ Type │ Source │ Target │ Status │ ├────┼──────┼───────────────────────┼───────────────────────┼───────────┤ │ 1 │ ¶ │ Hello, world! │ │ ❌ │ │ 2 │ ¶ │ This is a test. │ │ ❌ │ │ 3 │ ¶ │ Translate me! │ │ ❌ │ └────┴──────┴───────────────────────┴───────────────────────┴───────────┘ ``` **TMs** and **Termbases** are their own top tabs for managing translation memories and terminology. ### Status Icons | Icon | Meaning | | ---- | -------------------------------- | | ❌ | Not started | | ⚡ | Pre-translated | | ✏️ | Draft (edited but not confirmed) | | ✔ | Confirmed | | 🔒 | Locked | See [Segment Statuses](/workbench/editor/segment-statuses/) for the full list. ## Translating Your First Segment 1. Click on segment 1’s Target cell 2. Type your translation 3. Press `Ctrl+Enter` to confirm 4. The status changes to ✅ ### Using AI Translation 1. Click on segment 2 2. Press `Ctrl+T` to translate it with AI 3. The AI translation appears in the Target cell 4. Review, edit if needed, and confirm with `Ctrl+Enter` ## Setting Up Resources ### Add a Translation Memory 1. Go to the **TMs** tab 2. Click **+ Create TM** or **Import TMX** 3. Your TM will automatically provide matches ### Add a Termbase 1. Go to the **Termbases** tab 2. Click **+ Create Termbase** 3. Add terms manually or import from TSV ## Saving Your Project 1. Press `Ctrl+S` 2. Choose a name and location 3. Your project is saved as a folder containing the `.svproj` file, a `source/` folder (your original document), and – once you export – a `target/` folder. See [The Project Folder](/workbench/import-export/project-folder/). Tip **Tip:** Supervertaler auto-saves your work periodically, but it’s good practice to save manually before closing. To move a project, move the **whole folder**, not just the `.svproj`. ## Exporting the Translation When you’re finished: 1. Go to **Project → Export** 2. Choose your format: * **DOCX** - Standard Word document with translations * **Bilingual Table** - Source and target side by side * **Text File** - Plain text output 3. Select destination and click **Export** ## Project Workflow Summary ```plaintext Import Document ↓ Set Up TMs & Termbases (optional) ↓ Translate Segments (manual or AI) ↓ Review & Confirm (Ctrl+Enter) ↓ Save Project (.svproj) ↓ Export Translation ``` *** ## What’s Next? Now that you’ve completed your first project: * [Learn keyboard shortcuts](/workbench/editor/keyboard-shortcuts/) for faster work * [Set up batch translation](/workbench/ai-translation/batch-translation/) for larger documents * [Explore CAT tool workflows](/workbench/cat-tools/overview/) if you use professional tools # Installation ## Windows (Recommended) ### Option 1: Download Release (Easiest) 1. Go to [GitHub Releases](https://github.com/Supervertaler/Supervertaler-Workbench/releases) 2. Download the latest `.zip` file 3. Extract to a folder of your choice 4. Run `Supervertaler.exe` 5. *Optional:* double-click **`Add Supervertaler to Start Menu.cmd`** once to add a Start Menu shortcut, so you can launch the app from the Start Menu (or pin it to the taskbar) like any installed program. This is just a friendly wrapper around `create_start_menu_shortcut.ps1` that bypasses Windows’ default PowerShell ExecutionPolicy without changing any system-wide settings. ### Option 2: Run from Source If you want the latest development version or want to contribute: ```bash # Clone the repository git clone https://github.com/Supervertaler/Supervertaler-Workbench.git cd Supervertaler # Create virtual environment python -m venv venv venv\Scripts\activate # Install dependencies pip install -r requirements.txt # Run the application python Supervertaler.py ``` ## macOS The macOS install method depends on your Mac’s processor. ### Apple Silicon (M1, M2, M3, M4) – Download Release 1. Go to [GitHub Releases](https://github.com/Supervertaler/Supervertaler-Workbench/releases) 2. Download the latest `.dmg` file 3. Open the `.dmg` and drag **Supervertaler** to your Applications folder 4. Launch from Spotlight or Launchpad ### Intel Macs – Install via pip The published macOS `.dmg` is built for Apple Silicon only and will not run on Intel hardware. Intel Mac users need to install via pip and provide a system Java for the Okapi sidecar (which handles Word, Excel, HTML and other office-document imports). ```bash # 1. Install Homebrew (skip if already installed) /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" # 2. Install Python 3 (skip if already installed) brew install python@3.12 # 3. Install Java 17 (Eclipse Temurin, free, no Oracle account required) brew install --cask temurin@17 # 4. Install Supervertaler pip3 install supervertaler # 5. Run it supervertaler ``` If you skip the Java step, Supervertaler shows a friendly dialog at startup with the install command. Plain-text translation, TMX, termbases, etc. all work without Java – only office-document import/export needs it. On first DOCX import, Supervertaler downloads the Okapi sidecar JAR (\~28 MB) into `~/Library/Application Support/Supervertaler/okapi-sidecar/`. After that, everything runs locally and offline. ### Run from Source (any Mac) Follow the [Linux source instructions](#linux) below – the commands are identical except for the spellcheck step, where you’d use `brew install hunspell` instead of `apt install hunspell-*`. ## Linux Supervertaler is compatible with Linux, though Windows is the primary development platform. ```bash # Clone the repository git clone https://github.com/Supervertaler/Supervertaler-Workbench.git cd Supervertaler # Create virtual environment python3 -m venv venv source venv/bin/activate # Install dependencies pip install -r requirements.txt # Install Hunspell dictionaries (for spellcheck) sudo apt install hunspell-en-us hunspell-nl # Add your languages # Run the application python Supervertaler.py ``` Note **Linux Users:** If you experience crashes related to spellcheck or ChromaDB, see [Linux-Specific Issues](/workbench/troubleshooting/linux/). ## Dependencies The main dependencies are automatically installed via `requirements.txt`: | Package | Purpose | | ------------------- | ---------------------------- | | PyQt6 | User interface | | openai | OpenAI GPT integration | | anthropic | Anthropic Claude integration | | google-generativeai | Google Gemini integration | | python-docx | DOCX file handling | | pyspellchecker | Spellcheck | ## Next Steps After installation: 1. [Set up your API keys](/workbench/get-started/api-keys/) for AI translation 2. Follow the [Quick Start Guide](/workbench/get-started/quick-start/) 3. Create your [first translation project](/workbench/get-started/first-project/) # Quick Start Guide This guide will get you translating in under 5 minutes. If you’re not sure where to begin: import a file, translate a few segments, then export back to your CAT tool. ## Step 1: Start Supervertaler Launch the application by running `Supervertaler.exe` (Windows) or `python Supervertaler.py` (from source). Note If you want to use AI translation, set up your API keys first: [Setting Up API Keys](/workbench/get-started/api-keys/). ## Step 2: Import a Document 1. Go to **Project → Import** 2. Choose your file type: * **DOCX** - Standard Word documents * **Text File** - Plain text (one segment per line) * **memoQ Bilingual** - memoQ XLIFF or bilingual DOCX * **Trados Package** - SDLPPX files * **Phrase Bilingual** - Memsource bilingual DOCX * **CafeTran Bilingual** - CafeTran external view 3. Select source and target languages when prompted 4. Your document appears in the translation grid Note If you’re working with memoQ/Trados/Phrase/CafeTran, always choose the matching import option so tags and statuses round-trip correctly. ## Step 3: Navigate the Grid The translation grid has 5 columns: | Column | Description | | ---------- | --------------------------------------- | | **#** | Segment number | | **Type** | Segment type (¶, heading, list item, …) | | **Source** | Original text (read-only) | | **Target** | Your translation (editable) | | **Status** | Translation status indicator | ### Basic Navigation | Action | Shortcut | | ---------------- | ------------------------------- | | Next segment | `Enter` or `↓` (at end of cell) | | Previous segment | `↑` (at start of cell) | | Go to segment | `Ctrl+G` | Most navigation is designed to feel memoQ-like: arrow keys move within a cell, and at the top/bottom line they can jump between segments. ## Step 4: Translate a Segment ### Manual Translation 1. Click in the **Target** cell 2. Type your translation 3. Confirm the segment (confirmed statuses matter when exporting back to CAT tools) ### AI Translation 1. Select a segment 2. Use the **Translate action** (single segment; **Ctrl+T**) or **Batch Translate** (multiple segments) 3. Review and edit if needed 4. Confirm the segment when you’re happy Note If AI translation isn’t available yet, double-check provider setup in [Setting Up API Keys](/workbench/get-started/api-keys/). ## Step 5: Save Your Project 1. Press `Ctrl+S` or go to **Project → Save Project** 2. Choose a location and filename 3. Projects are saved as `.svproj` files ## Step 6: Export Your Translation 1. Go to **Project → Export** 2. Choose the appropriate format: * **DOCX** - Translated Word document * **Bilingual Table** - Side-by-side source/target * **Return Package** - For CAT tool workflows Caution For CAT tool workflows, always export the matching return format (for example, SDLRPX for Trados return packages) to preserve tags and statuses. *** ## What’s Next? ### Recommended next steps * [Setting Up API Keys](/workbench/get-started/api-keys/) – enable AI translation * [Installation](/workbench/get-started/installation/) – verify dependencies and optional components * [CAT Tool Integration](/workbench/cat-tools/overview/) – memoQ/Trados/Phrase/CafeTran workflows * [Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) – faster editing and navigation | | | | ---------------------------- | ------------------------------------------------------- | | **Set up AI Translation** | [Configure API keys →](api-keys.md) | | **Learn Keyboard Shortcuts** | [View all shortcuts →](../editor/keyboard-shortcuts.md) | | **Work with CAT Tools** | [CAT tool integration →](../cat-tools/overview.md) | # Supervertaler Re-importable Table (DOCX) The **Supervertaler Re-importable Table** is a branded Word (DOCX) export that lays your project out as a side-by-side table – handy for reviewing, proofreading, or handing off to someone who doesn’t use a CAT tool. Its defining feature: it can be edited and **re-imported** to pull the changes straight back into your project. (Its plain-text sibling, [Re-importable Text](/workbench/import-export/bilingual-text/), does the same round-trip in an AI-friendly text format.) ## When to use it * A proofreading round-trip in Word: export, edit the targets, re-import. * Sharing with a reviewer who doesn’t use your CAT tool. ## Columns | Column | Description | | ------------ | ----------------------------------------------------------------------- | | **#** | Segment number | | **Source** | Source text | | **Target** | Target text (edit this when proofreading) | | **Status** | Segment status | | **Comments** | Segment comments – edit, add, or clear; changes round-trip on re-import | A header above the table shows the project name, language pair, segment count, and export date. ## Where to find it **Project → Export → 🔁 Supervertaler Re-importable → Bilingual Table (DOCX)**. Formatting tags stay visible as markup; edit the Target/Comments cells, save, and bring the changes back in (see below). Don’t change the segment numbers (#) or the source text. The document is titled *Supervertaler Re-importable Table*.  The re-importable Bilingual Table: formatting shown as visible markup, with a notice that segment numbers and source text must stay unchanged so the file can be re-imported after proofreading. Note Need a clean copy with real bold/italic for a client, rather than visible tags? Export the finished document itself via **Project → Export → Export Translated Document** – it renders formatting properly in the original file layout. ## Round-trip (proofread and re-import) 1. Export the **re-importable** Bilingual Table. 2. Edit in Word – leave the **#** and **Source** columns untouched: * The **Target** column for translation edits. * The **Comments** column to edit, add, or clear segment comments. New comments added to segments that had none in Workbench are also round-tripped. 3. Back in Workbench: **Project → Import → 🔁 Supervertaler Re-importable → Bilingual Table (DOCX) – Update Project**. 4. Supervertaler diffs the file against your project and shows a preview before applying. Target changes set the segment back to “Not Started” so you can re-confirm; comment changes replace the segment’s existing comments verbatim with the proofreader’s text (no `[Review: …]` wrapping or appending – round-trip in, round-trip out). Note Re-imports written by Workbench v1.10.182 and earlier had a bug where comments-only edits were silently discarded – only segments whose target text *also* changed had their comments updated. Fixed in v1.10.183: a comment edit on its own is now picked up, and a comment cleared in the bilingual file clears it on the segment. ## Other bilingual tables To round-trip back into a **CAT tool**, use that tool’s own bilingual format (memoQ, CafeTran, Phrase, Trados Bilingual Review) rather than the Supervertaler table – see [CAT Tool Overview](/workbench/cat-tools/overview/). ## Related * [Supported File Formats](/workbench/import-export/formats/) * [Exporting Translations](/workbench/import-export/exporting/) # Supervertaler Re-importable Text (AI-friendly) The **Re-importable Text** round-trip lets you send a whole translation out as a plain-text file – for a proofreader or an LLM to edit – and then pull the edits straight back into the same project. It’s the plain-text sibling of the [Re-importable Table (DOCX)](/workbench/import-export/bilingual-tables/), ported from the Supervertaler for Trados plugin. Added in v1.10.231. Note **Why “Text” and not “Markdown”?** The file is deliberately plain text. Its segment blocks rely on line breaks being preserved, and a Markdown renderer collapses single line breaks – which would scramble the structure. AI agents read the raw characters when you paste a file into a chat, so plain text is both safe and maximally readable. ## Exporting **Project → Export → 🔁 Supervertaler Re-importable → Bilingual Text (AI-friendly)…** You’ll get a small options dialog (include locked segments; which statuses to include), then a save dialog. Two files are written side by side: * `MyProject_bilingual.txt` – the editable text file. * `MyProject_bilingual.txt.svexport.json` – a **sidecar** that records, per segment, a stable id, a source hash, and the status. Keep the two files together; the sidecar is what makes a safe re-import possible. The file opens with a short header that lists exactly which statuses you may set. Each segment is one block: ```plaintext [SEGMENT 0001] EN: The quick brown fox {1} NL: De snelle bruine vos {1} Status: Confirmed Comment: Verify the shade of "brown" ``` * The `EN:` line is the **source** – leave it alone. It stays on **one line**, and a `[newline]` in it marks where the original source broke across two lines (e.g. a subtitle cue). The source is read-only and never written back to your project, so these tokens are just there to show its structure – handy for spotting a target that’s missing a break the source has. * The `NL:` line is the **target** – edit it freely, but **keep it on one line**. Where the target needs a hard line break – for example to split a subtitle across two lines – write the literal token `[newline]`: ```plaintext NL: Welkom bij dit webinar[newline]over de waardeketenanalyse ``` On re-import `[newline]` is turned back into a real line break, so the two-line layout is preserved on export. *(Introduced in v1.10.255; files exported before that – with the target genuinely wrapped over several lines – still re-import unchanged.)* * The `Comment:` line is always present (blank when the segment has no comment) so you can see the field exists. **Edit it, fill the blank one, or clear it** – the change re-imports into the segment’s comments. It too may span several lines. * `…`, `…`, `…` are **cosmetic formatting** – add or remove them as you like. * `{1}`, `[1}`, `<92>` and similar are **structural tags** – keep them; dropping one will flag the segment on re-import (see below). A real file looks like this – note the single-line sources and the `[newline]` token marking where each target splits across two lines: ![A Supervertaler Re-importable Text file: the header lists the project details and the editing rules, the NL: source lines each sit on one line, and two EN: targets show the \[newline\] token highlighted where a subtitle is split across two lines.](/.gitbook/assets/SUPERVERTALER-RE-IMPORTABLE-TEXT-newlines.png) ## Editing with an LLM Hand the `.txt` to ChatGPT, Claude, Gemini, etc. with an instruction such as *“edit only the `NL:` lines; keep each one on a single line, using the literal token `[newline]` for any line break; leave the `[SEGMENT …]` markers, the `EN:` lines, and the `{…}` tags untouched.”* Because each field is one labelled line (not a table column), it survives pipe characters and long inputs without the source/target roles getting confused, and keeping targets to one line stops an agent from accidentally reflowing them. ## Re-importing **Project → Import → 🔁 Supervertaler Re-importable → Bilingual Text (AI-friendly) - Update Project…** Pick the edited `.txt` (its sidecar is found automatically). A preview dialog shows how many segments will be updated, how many are unchanged, and how many are skipped – and why. Nothing is applied until you click **Apply changes**. ### Safety guards * **Source-tamper detection** – if a segment’s source line was changed, that segment is skipped (its hash no longer matches the sidecar). * **Structural-tag integrity** – if the edited target dropped a required tag, the segment is flagged. With **“Refuse to apply edits that drop required tags”** ticked (the default), such segments are skipped; untick it to apply them anyway. Cosmetic ``/``/`` changes never trip this. * **Locked segments** are never modified. ### Status * If you (or the AI) deliberately change a `Status:` line to a different value, that status is applied. * Otherwise, any segment whose target you edited is marked **Draft** – a translated-but-unconfirmed state, ready for you to review and confirm. ### Comments * Existing segment comments are written out as a `Comment:` line. Edit it, add a new one to a segment that had none, or delete it – the change re-imports into the segment’s comments. A comment-only edit (target left alone) is applied on its own and shows up in the import preview. ## Tips * Export reads **live grid state**, so in-progress edits are included even if the segment isn’t confirmed. * If the sidecar is missing, you can still re-import – segments are then matched by position only, and source-tamper detection is unavailable. You’ll be warned first. ## Related * [Supervertaler Re-importable Table (DOCX)](/workbench/import-export/bilingual-tables/) * [Exporting Translations](/workbench/import-export/exporting/) # Importing DOCX Files Use DOCX import for normal Word documents (not CAT tool bilingual formats). DOCX import is handled by the bundled **Okapi sidecar** – an industry-standard localisation library that runs as a small background service. This gives you SRX-based segmentation, proper paragraph and table detection, and a faithful round-trip on export. ## Import steps 1. Go to **Project → Import → Import Document…** (`Ctrl+O`) 2. Select your `.docx` file 3. Choose the **source** and **target** languages when prompted 4. The document is segmented into rows in the translation grid – formatting tags (``, ``, ``, hyperlinks, runs) appear inline so you can preserve them in the translation A progress dialogue shows extraction progress for large documents – on a 2,500-segment file expect a few seconds. ## What is preserved on round-trip When you later export back to DOCX, Supervertaler reconstructs the original document via the same Okapi sidecar: * **Layout**: paragraphs, tables, headers, footers, page breaks * **Inline formatting**: bold, italic, underline, sub/superscript, colour * **Hyperlinks**: anchor text and links round-trip identically – broken or working * **Images and other non-translatable content**: passed through untouched Note The document you import becomes the project’s **source**. When you save the project, Supervertaler copies it into the project’s `source/` folder and refers to it by a relative path – so this faithful round-trip keeps working even if you later move, rename, or delete the original you imported from. See [The Project Folder](/workbench/import-export/project-folder/). ## Tips * If you are working with memoQ/Trados/Phrase/CafeTran, prefer the specific CAT workflow import instead of generic DOCX. * Keep formatting tags balanced when editing translations (e.g. `text`, not `text`). * Placeholder tags like `` and `` represent structural elements (hyperlinks, run boundaries). Leave them in the same position as the source so the export reconstructs the document correctly. ## When DOCX import is unavailable The Okapi sidecar requires Java – Supervertaler ships a bundled JRE, so this is normally invisible. If the sidecar fails to start (Java missing, port 8090 blocked by another process, …) you’ll see an “Okapi sidecar required” dialogue with troubleshooting steps. DOCX import won’t fall back silently to a degraded engine. Note Need OCR? Use [PDF Rescue (OCR)](/workbench/tools/pdf-rescue/) to turn scanned PDFs into editable DOCX before importing. # Export Verification (Word-Count Check) Whenever you export a translated **DOCX**, Supervertaler runs an automatic safeguard that checks whether any text was lost on the way out of the document. It is a safety net against the rare case where the round-trip drops content that is present and confirmed in the grid. ## What it does After writing the file, Supervertaler: 1. Counts the words it **expected** to write – the target text of every segment, falling back to the source text for any untranslated segment. 2. Counts the words **actually present** in the exported DOCX (document body, headers, footers, and foot/endnotes). 3. Compares the two. If the exported file contains noticeably fewer words than expected, it shows a warning. Tags, numbers, and punctuation are counted the same way on both sides, so a genuine loss of text shows up as a clear shortfall while incidental formatting differences stay within tolerance. ## When it runs * Automatically, on **every DOCX export** – single-file and multi-file, whether the file is built through the Okapi merge or the standard exporter. * Multi-file projects produce a **single combined warning** listing each affected file, rather than one dialog per file. ## The warning If a file falls short, you’ll see a **Possible Missing Text in Export** dialog naming the file(s) and roughly how many words appear to be missing. The same result is written to the log, for example: ```plaintext 🔢 Export word-count check [Manual.docx]: 4065/4060 words (100%; threshold 95%) ``` or, when text looks lost: ```plaintext ⚠️ Possible dropped text in export: Manual.docx has only 85% of the expected words – review before delivery. ``` When you see the warning, open the file and check it before delivering. Note The check is deliberately **coarse**. It reliably catches a material loss (for example a whole paragraph or many segments going missing), but a single very short segment dropping out can stay within the tolerance band and won’t trigger a warning. It’s a backstop, not a substitute for a final read-through. ## Adjusting or turning it off The check is configured in your `settings.json` file, under an `"export"` section: | Key | Default | Meaning | | ---------------------------- | ------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `word_count_check_enabled` | `true` | Set to `false` to turn the check off entirely. | | `word_count_check_threshold` | `0.95` | Warn when the exported file has fewer than this fraction of the expected words. Raise it (e.g. `0.99`) for more sensitivity, lower it to tolerate larger differences. | ```json { "export": { "word_count_check_enabled": true, "word_count_check_threshold": 0.95 } } ``` If your documents legitimately differ a lot from the segment word count – for example they contain many numbers, or comments that aren’t part of the translation – you may prefer to lower the threshold slightly to avoid false alarms. Caution The check currently applies to **DOCX exports only**. Other Okapi formats (IDML, HTML, XLIFF, PPTX, XLSX, PO) are not yet verified this way. ## Related pages * [Exporting Translations](/workbench/import-export/exporting/) * [Multi-File Projects](/workbench/import-export/multi-file/) * [Supported File Formats](/workbench/import-export/formats/) # Exporting Translations When you’re done translating, export in a format that matches your workflow. ## Export steps 1. Go to **Project → Export** 2. Choose an export type (for example DOCX, bilingual table, or a CAT return format) 3. Pick a destination and save Tip For **Export Translated Document** (and Simple Text), the Save dialog opens in your project’s `target/` folder by default, so finished translations land alongside their sources. You can still browse elsewhere – see [The Project Folder](/workbench/import-export/project-folder/). ## CAT tool round-trips If you started from a CAT exchange format (memoQ/Trados/Phrase/CafeTran), export the matching return format. ### Important rules for round-trips * **Segment count must match**: don’t merge or split segments. * **Keep tags balanced**: for example `text` (not `text`). * **Don’t “pretty edit” bilingual tables**: changing the table structure in Word can break reimport. * Run your CAT tool’s QA after reimport. Caution Don’t merge or split segments in Supervertaler when you plan to reimport into a CAT tool. ## Choosing the right export * For CAT tool workflows, use the matching CAT export: * memoQ bilingual DOCX * Trados return package (SDLRPX) when you imported SDLPPX * Phrase bilingual DOCX * CafeTran bilingual table DOCX * For review-only delivery, consider [Bilingual Tables](/workbench/import-export/bilingual-tables/). ## Checking the export After every DOCX export, Supervertaler automatically compares the word count of the exported file against your translated segments and warns you if text looks like it was dropped. See [Export Verification (Word-Count Check)](/workbench/import-export/export-verification/). ## Related pages * [Supported File Formats](/workbench/import-export/formats/) * [Export Verification (Word-Count Check)](/workbench/import-export/export-verification/) * [Bilingual Tables](/workbench/import-export/bilingual-tables/) * [CAT Tool Overview](/workbench/cat-tools/overview/) # Supported File Formats Supervertaler can import and export several formats depending on your workflow. ## Standard documents * **DOCX** (Microsoft Word): import a document, translate in the grid, export a translated DOCX. * **TXT** (plain text): each line becomes a segment. ## Other formats via Okapi The bundled **Okapi sidecar** lets Supervertaler round-trip a wider set of formats. Pick a file in any of the following types via **Project → Import → Import Document…**, translate in the grid, then **Project → Export → Export Translated Document…** to write a translated file back in the same format: | Format | Extension(s) | Notes | | ------------------------- | ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | **Adobe InDesign Markup** | `.idml` | Drop in, translate, drop out – no need to round-trip via Trados/memoQ first. Inline tags appear as `...` markers in the grid; preserve them in the translation. | | **HTML** | `.html`, `.htm` | Anchors, images, buttons, and other inline elements are exposed as `...` / `` tags. The translated HTML reconstructs the original markup byte-perfectly. | | **XLIFF 1.2** | `.xliff`, `.xlf` | The industry-standard bilingual interchange format. Useful for files exported from any CAT tool that doesn’t have its own dedicated entry. | | **gettext PO** | `.po` | Source strings are translated; `msgctxt` and plural forms are preserved. | | **Microsoft Excel** | `.xlsx` | Cells, formulas, and styling round-trip via the Office Open XML filter. | | **Microsoft PowerPoint** | `.pptx` | Slides and slide notes are extracted; layout and master slides round-trip. | ### How it works Okapi extracts the translatable content from the source file plus a *skeleton* file that preserves the original structure. You translate the extracted content; the merge step combines your translation with the skeleton to reconstruct the original format with the new text in place. Note **Tag handling**: when you see `` / `` / `` markers in the source segment, leave them in the translation in the same positions. They map back to inline elements like links, buttons, or formatting runs in the original file. ## CAT tool exchange formats Use these formats when you need to round-trip back into a CAT tool. * **memoQ** * Bilingual DOCX * XLIFF (memoQ export) * **Trados Studio** * Packages: `.sdlppx` import → `.sdlrpx` return (recommended) * Bilingual Review DOCX (special workflow) * **Phrase (Memsource)** * Bilingual DOCX * **CafeTran Espresso** * Bilingual DOCX table ## Multi-file projects * **Folder import (Multiple Files)**: import a folder containing DOCX/TXT files into a single multi-file project. Caution For CAT tool round-trips, always import and export the matching CAT format. Mixing formats can break tags/statuses on reimport. ## Related pages * [Importing DOCX Files](/workbench/import-export/docx-import/) * [Importing Text Files](/workbench/import-export/txt-import/) * [Multi-File Projects](/workbench/import-export/multi-file/) * [Exporting Translations](/workbench/import-export/exporting/) * [Bilingual Tables](/workbench/import-export/bilingual-tables/) # Import Options (File Types) When you import an Office document (Word, Excel or PowerPoint), Supervertaler uses the bundled **Okapi sidecar** to decide which parts of the file become translatable segments. The **File Types** options let you control that – so you can pull in (or leave out) things like comments, hidden text, headers and footers, and speaker notes. ## Where to set them There are two places, and they work together: * **Settings → 📄 File Types** – your **defaults**, applied to every import. Tick a box, and it’s saved straight away. Use **Restore defaults** to go back to the recommended set. * **The import dialog** – when you import a single document or a folder, the same options appear there, pre-filled from your defaults. Any change you make applies to **that import only**, overriding the default without changing it. Note These options apply to Okapi-based imports (DOCX, XLSX, PPTX). Plain text and Markdown files, and CAT-tool bilingual formats, are unaffected. ## Word (DOCX) | Option | What it does | | ------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Comments** | How to handle Word review comments. **Skip** (default) leaves them out. **Import as comments** brings them in as real comments – anchored to the relevant segment where possible (otherwise to the top of the document), tagged with the original reviewer’s name so they’re easy to tell from your own. These are shown for context and are *not* re-exported (the originals stay in the file). **Import as translatable text** brings each comment in as a segment to translate (labelled “Cmt” in the Type column); those *are* written back on export. | | **Import hidden text** | Include text formatted as hidden. Off by default. | | **Import headers & footers** | Include page headers and footers. On by default. | | **Import document properties** | Include title, author, keywords and similar metadata. Off by default. | | **Skip drawing/shape names** | Leave out the auto-generated object names Word gives every shape and group (for example *Shape 16*, *Group 574248*). On by default – these are not real content and would otherwise clutter the grid. | | **Accept tracked changes** | Use the final, accepted text of any tracked changes. On by default. | ## Excel (XLSX) | Option | What it does | | ------------------------------------- | --------------------------------------------------------------- | | **Import hidden rows/columns/sheets** | Include content that is hidden in the workbook. Off by default. | | **Import sheet (tab) names** | Include the worksheet tab names. Off by default. | | **Import text in shapes/text boxes** | Include text drawn on the sheet. On by default. | ## PowerPoint (PPTX) | Option | What it does | | ---------------------------------- | ----------------------------------------------------- | | **Import speaker notes** | Include the notes beneath each slide. Off by default. | | **Import comments** | Include review comments. Off by default. | | **Import hidden slides** | Include slides marked hidden. Off by default. | | **Import slide masters / layouts** | Include text on masters and layouts. On by default. | ## How imported parts are labelled Anything that isn’t ordinary body text is tagged in the grid’s **Type** column so you can tell at a glance where it came from: * **Cmt** – a comment * **Hdr** / **Ftr** – a header or footer * **Prop** – a document property * **Note** – a PowerPoint speaker note ## Round-trip on export The options you import with are remembered with the project and reused when you export. That keeps the document structure aligned, so the translated file comes back out cleanly – there’s nothing extra to set at export time. # Multi-File Projects Multi-file projects let you import a whole folder of files as a single Supervertaler project. ## Import a folder 1. Go to **Project → Import → Folder (Multiple Files)…** 2. Choose a folder containing supported files (DOCX/TXT/MD) 3. Select which files to include 4. Choose the **source** and **target** languages ## How it behaves * All segments live in one grid, but each segment is associated with a source file. * You can jump between files and track progress per file. * Each DOCX file is imported via the **Okapi sidecar** (the same engine that handles single-file DOCX import) – you get SRX segmentation, faithful round-trip, and hyperlinks/structural tags preserved per file. * TXT and MD files use the simple per-line import. * Original files are backed up to a `_source_files/` folder inside the project folder so the export can reconstruct each document faithfully. ## Export When you export a multi-file project: * Each DOCX file is reconstructed via the Okapi sidecar’s `/merge` endpoint, using the original from `_source_files/` as the template. Layout, formatting, hyperlinks, and tables round-trip identically. * TXT/MD files are written with the same per-line structure as the source. * Output files land in the destination folder you choose, named `_translated.`. ## Tips * Use this when you receive a set of related files (e.g. claim documents, manual chapters split into separate files, UI strings split across documents). * Export is done in one operation – pick the destination folder and Supervertaler writes all the translated files at once. ## Requirements * The Okapi sidecar must be running before you import any folder containing DOCX files. Supervertaler checks this up-front and shows an “Okapi sidecar required” dialogue if it can’t reach the sidecar – better than failing halfway through importing twenty files. # The Project Folder A Supervertaler project isn’t just the `.svproj` file – it’s a **folder** that holds the project file together with the documents it works on. Keeping everything in one folder means a project is self-contained: you can move, rename, zip or email the folder and it still opens and exports correctly. Note **New Project** has a ”📁 Create a dedicated folder for this project” checkbox (on by default). When it’s on, the first save tucks the `.svproj` into its own folder as shown below. Turn it off to save the `.svproj` flat, wherever you choose – the folder layout is never forced, so you can keep your own naming or nest a project inside a larger job folder. ## What’s in a project folder ```plaintext My Project/ ├─ My Project.svproj ← the project file ├─ source/ ← the original documents you're translating └─ target/ ← the translated documents you export ``` * **`source/`** – when you **save** a project, its original document is copied here, and the project remembers it by a path *relative* to the folder. That’s what makes the project portable: it no longer depends on the document staying at the exact location you first imported it from. Move or rename the original afterwards and your export still works. * **`target/`** – when you run **Project → Export → Export Translated Document** (or Simple Text), the Save dialog opens here by default, so your finished translations land next to their sources. You can still browse somewhere else; this is only the default. ## Why this matters * **Portability** – hand the whole folder to a colleague, or move it between machines, and the structure-preserving export keeps working. Nothing points at a file that only exists on your computer. * **No accidental cross-wiring** – because the source is stored relative to the project folder, a project can never end up bound to an unrelated document. Tip Keep the `.svproj` **inside** its folder. If you want to relocate a project, move or copy the **whole folder**, not just the `.svproj` on its own. ## Existing projects Projects created before this layout existed still work – they reference their source by an absolute path, and Supervertaler resolves it as before. The next time you **save** such a project, its source is copied into `source/` and the reference switches to the portable relative form automatically. ## Related pages * [Exporting Translations](/workbench/import-export/exporting/) * [Your First Translation Project](/workbench/get-started/first-project/) * [Multi-File Projects](/workbench/import-export/multi-file/) # Pseudo-translation (Export Test) **Pseudo-translation** fills your targets with deliberately stress-tested placeholder text so you can export the document and check that it comes out with correct formatting, layout, fonts and tags – **before** you invest any time in real translation. It’s a pre-flight check borrowed from desktop CAT tools. Find it under **Bulk Operations → 🧪 Pseudo-translate (Export Test)…**. ## Why not just copy source to target? Copying source into target and exporting tests the *plumbing* – does the file merge and export, do the inline tags survive – but it misses the problems that actually bite at the end of a job: * **Length stays identical**, so overflowing text boxes, clipped table cells, fixed-width fields, reflow and truncation never show up. Real translations change length. * **Characters are never exercised** – the source already renders fine in the document’s fonts and encoding, so copying it tells you nothing about whether the *target* language’s characters will. * **Dropped or merged segments stay invisible**, because target text that equals the source still looks correct. Pseudo-translation addresses all three at once. ## What it does For every segment in the chosen scope it rewrites the target so that: 1. **Inline tags are preserved exactly.** Formatting tags (``, ``, …), Trados/SDLXLIFF numeric tags (`<410>`) and memoQ tags are kept verbatim and in order – only the words between them are changed. This is what makes the test trustworthy: the tag round-trip you’re checking isn’t disturbed. 2. **The text is length-expanded** by a ratio you choose, to surface overflow and layout breaks. 3. **Characters are optionally accented** (`werkwijze` → `wéřkwíjžé`) to test diacritics, encoding and font coverage. 4. **Each segment is wrapped in `⟦ ⟧` markers** so a dropped, merged or misplaced segment is obvious in the exported file. A segment like ```plaintext De uitvinding betreft een werkwijze voor het sorteren. ``` becomes something like ```plaintext ⟦Dé úítvíñdíñğ bétřéft ééñ lorem wéřkwíjžé lorem vóóř hét šóřtéřéñ. lorem⟧ ``` ## Options | Option | What it controls | | -------------------- | --------------------------------------------------------------------------- | | **Apply to** | All segments (default), the filtered/visible set, or just your selection. | | **Length expansion** | 0% (structure only), +30% (typical), up to +200% (max stress). | | **Characters** | *Accented* (tests encoding/fonts) or *Plain words* (length + markers only). | | **Boundary markers** | Wrap each segment in `⟦ ⟧`. On by default. | ## Workflow 1. Open the project and run **Bulk Operations → 🧪 Pseudo-translate (Export Test)…**. 2. Pick your options and confirm. 3. Export the document the way you normally would (**Project → Export → Export Translated Document**, or any bilingual/CAT export) and open the result. 4. Check for clipped text, broken tables, missing glyphs, reordered or missing `⟦ ⟧`-marked segments, or tag errors. 5. **Edit → Undo** restores your real (usually empty) targets – the whole operation is recorded as a single reversible step. Note Pseudo content overwrites existing targets (Undo restores them). If you’d rather keep your working project untouched, run the test on a **copy** of the project. ## Related * [Export Verification (Word-Count Check)](/workbench/import-export/export-verification/) * [Exporting Translations](/workbench/import-export/exporting/) * [Re-importable Table (DOCX)](/workbench/import-export/bilingual-tables/) # Importing Text Files Text import is the simplest workflow: **each line becomes one segment**. ## Import steps 1. Go to **Project → Import → Text / Markdown File (TXT, MD)…** 2. Select your `.txt` file 3. Choose the **source** and **target** languages ## Tips * Keep one sentence (or one logical unit) per line for best results. * If your file has encoding issues (weird characters), try saving it as UTF-8. ## Export Text projects can be exported as: * Translated TXT * DOCX * Bilingual table # Non-Translatables Non-translatables are terms you want to keep unchanged (product names, IDs, codes, etc.). ## What you can do * Maintain a list of non-translatable terms * Highlight them in the grid * Reduce accidental changes during editing ## Tips * Add non-translatables before batch translation for best results. * Use them for brand names, UI strings, and technical identifiers. # AI Proofreading **AI Proofreading** asks an LLM to review your finished translation for accuracy, completeness, terminology and style, and records what it finds as **proofreading comments** on each segment. It’s a translator-side review pass – nothing is changed automatically; you read the feedback and decide what to act on. Proofreading lives under the top-level **QA** menu: * **QA ▸ Proofreading ▸ Proofread Translation…** – run a proofreading pass. * **QA ▸ Proofreading ▸ Delete All Proofreading Comments** – clear every proofreading comment in the project. ## Running a proofreading pass Open **QA ▸ Proofreading ▸ Proofread Translation…**. The dialog has three things to set: ### 1. Which segments | Scope | What it checks | | -------------------------------- | ------------------------------------------------------------------------ | | **✅ Confirmed only** *(default)* | Only segments you’ve confirmed. | | **📝 Translated + Confirmed** | Draft *and* confirmed segments. | | **🔹 Selected** | The rows you’ve selected in the grid (select rows first to enable this). | | **🌐 All segments** | Every segment, regardless of status. | ### 2. Which model Proofreading uses your **currently-active AI provider and model** (set in **AI Settings**) – the dialog shows which one, e.g. `📊 Using: Openai (gpt-5.5)`. To proofread with a *different* model, switch the active provider in AI Settings and run the pass again (see [Multiple models](#multiple-models) below). ### 3. Which prompt * **Default (built-in)** runs the standard four-point check: 1. **Accuracy** – does the target correctly convey the source meaning? 2. **Completeness** – is anything missing or added? 3. **Terminology** – are technical terms correct and consistent? 4. **Grammar & Style** – is the text natural and error-free? * Or pick a **custom proofreading prompt** from the dropdown – any prompt you’ve saved under the **Bulk Operations/** folder of your [Prompt Library](/workbench/ai-translation/prompt-library/) appears here. * Or **type a one-off prompt** straight into the box. Click **Proofread** to start. A progress dialog shows how many segments have been checked, how many issues were found, and how many came back clean; you can cancel partway through. ## Where the results appear Findings land in the **✅ Proofreading** sub-tab of the **💬 Comments** panel – an all-project list of every proofreading comment, one entry per (segment, model). See [Comments → Proofreading comments](/workbench/editor/comments/#proofreading-comments) for the full rundown. In short: * Each entry has a clickable **Segment #N · model** header that jumps to the segment. * Selecting a segment in the grid **scrolls and highlights** the list to that segment’s comments. * A **🗑️** button deletes a single comment; **QA ▸ Proofreading ▸ Delete All Proofreading Comments** clears them all. * In the grid, a segment with a proofreading comment shows a **purple** Status-cell background (versus **amber** for a segment comment, and a **split** when it has both). ## Multiple models Results are stored **keyed by model**, so passes with different models *accumulate* rather than overwrite: proofread once with GPT and once with Claude, and each segment keeps both sets of findings. In the Proofreading list **each engine gets its own colour**, so you can compare at a glance what each model flagged. Running the same model again replaces only that model’s note. ## Good to know * **Proofreading comments are ephemeral review notes.** They’re stored in the `.svproj` project file but are **not exported** to your final document or bilingual tables – unlike [segment comments](/workbench/editor/comments/), which do export as Word comments. Deleting them is safe: another proofreading pass regenerates them. * Proofreading is **read-only feedback** – it never edits your target text for you. * Cost scales with scope and model. Proofreading every segment with a premium model on a large project is a real API spend; the **Confirmed only** default keeps a first pass focused. See [Usage & Costs](/workbench/ai-translation/usage-costs/). ## Related * [Comments](/workbench/editor/comments/) – where proofreading comments are listed and managed * [Spellcheck](/workbench/qa/spellcheck/) · [Tag Validation](/workbench/qa/tag-validation/) · [Non-Translatables](/workbench/qa/non-translatables/) * [Prompt Library](/workbench/ai-translation/prompt-library/) – save custom proofreading prompts * [Usage & Costs](/workbench/ai-translation/usage-costs/) # Spellcheck Supervertaler includes a powerful spellcheck system that highlights misspellings while you translate, with support for regional language variants. ## How It Works Supervertaler uses a **three-tier spellcheck system** that automatically selects the best available backend: | Backend | Description | Languages | | ------------------------- | ---------------------------------------------- | ------------------------------------------------- | | **Hunspell (cyhunspell)** | Native C library, best accuracy | Any language with .dic/.aff files | | **Spylls** | Pure Python Hunspell (recommended for Windows) | Bundled: EN, RU, SV + any .dic/.aff files you add | | **pyspellchecker** | Built-in fallback | EN, NL, DE, FR, ES, PT, IT, RU | The system automatically falls back through backends: Hunspell → Spylls → pyspellchecker. Note **Windows Users:** Spylls is automatically used since cyhunspell doesn’t compile on Python 3.12+. This works great and supports regional variants! ## Features * **Red wavy underlines** for misspelled words in the translation grid * **Right-click context menu** with spelling suggestions * **Add to Dictionary** – Save a word permanently * **Ignore** – Skip a word for the current session only * **Regional variants** – Distinguish between en\_US “color” and en\_GB “colour” ## Language Variants Supervertaler supports regional language variants. The spellcheck dropdown shows variants like: * English (US), English (GB), English (AU), English (CA), English (ZA) * Portuguese (PT), Portuguese (BR) * Spanish (ES), Spanish (MX), Spanish (AR) * French (FR), French (CA), French (BE) * German (DE), German (AT), German (CH) * Dutch (NL), Dutch (BE) Tip **Regional spelling works correctly!** * With **English (GB)**: “colour” ✅ correct, “color” ❌ incorrect * With **English (US)**: “colour” ❌ incorrect, “color” ✅ correct ## Spellcheck Info Dialog Access detailed information about your spellcheck setup: 1. Click the **🔤 Spellcheck** button in the grid toolbar 2. Or go to **View → Spellcheck Info** The dialog shows: * Current language and backend * Available languages * Diagnostic information (which backends are available/initialized) * Links to download additional dictionaries * Custom dictionary word count ## Adding More Dictionaries To add spellcheck support for additional languages or variants: 1. **Download Hunspell dictionaries** (.dic and .aff files) from: * [hunspell.memoq.com](https://hunspell.memoq.com/) – 70+ languages * [GitHub: wooorm/dictionaries](https://github.com/wooorm/dictionaries/tree/main/dictionaries) – 92+ languages * [LibreOffice Extensions](https://extensions.libreoffice.org/?Tags%5B%5D=50) – Rename .oxt to .zip 2. **Extract the files** – You need both `.dic` and `.aff` files (e.g., `nl_NL.dic` and `nl_NL.aff`) 3. **Place them in the dictionaries folder:** * Open Supervertaler * Go to Spellcheck Info dialog * Click ”📁 Open Dictionaries Folder” * Copy your .dic and .aff files there * You can also organize in subfolders (e.g., `dictionaries/en/en_GB.dic`) 4. **Restart Supervertaler** – The new language will appear in the dropdown Note **Spylls bundled dictionaries** (EN, RU, SV) are stored inside the spylls pip package, not in your dictionaries folder. Add your own .dic/.aff files to the dictionaries folder to extend available languages. ## Custom Dictionary You can add words that Supervertaler should always accept: * **Right-click a “misspelled” word** → **Add to Dictionary** * Or manually edit `user_data/dictionaries/custom_words.txt` Custom words are stored permanently and apply to all languages. ## Troubleshooting ### Spellcheck not working? 1. **Check the language** – Make sure the correct language variant is selected 2. **Check the backend** – Open Spellcheck Info to see which backend is active 3. **Missing dictionaries** – Some languages require manual dictionary installation ### Wrong language variant? If you need British English but only have US English: 1. Download `en_GB.dic` and `en_GB.aff` from one of the dictionary sources 2. Place them in your dictionaries folder 3. Select “English (GB)” from the dropdown ### Linux crashes? On Linux, some Hunspell configurations can cause crashes. Try: * Installing proper Hunspell dictionaries: `sudo apt install hunspell-` (e.g., `hunspell-pl` for Polish) * Temporarily disabling spellcheck in Settings → View Settings * See [Linux-Specific Issues](/workbench/troubleshooting/linux/) for more details ## Technical Details For developers and advanced users: | Project | Description | | ----------------------------------------------------------- | ----------------------------------- | | [pyspellchecker](https://github.com/barrust/pyspellchecker) | Built-in word frequency spellcheck | | [spylls](https://github.com/zverok/spylls) | Pure Python Hunspell implementation | | [Hunspell](http://hunspell.github.io/) | Original C/C++ spellcheck library | The spellcheck manager is located in `modules/spellcheck_manager.py` and provides: * Automatic backend selection * Dictionary file detection (including subdirectories) * Word caching for performance * Custom dictionary management # Tag Validation When working with formatted documents or CAT tool files, **tags must be preserved**. ## Why tags matter Tags represent formatting or placeholders. If tags are missing or unbalanced, reimporting into your CAT tool can fail or formatting may be lost. ## Tag display modes Supervertaler supports two ways of viewing formatting: * **WYSIWYG mode**: shows *bold/italic/underline* as formatting * **Tag view**: shows the raw markup (for example `...`) Use **Tag view** when you are preparing to export/reimport and you want to verify the raw tags. ## Supported formatting tags These tags are commonly used in Supervertaler projects: | Tag | Meaning | | ---------------- | ------------- | | `...` | Bold | | `...` | Italic | | `...` | Underline | | `...` | Bold + Italic | | `<sub>...</sub>` | Subscript | | `<sup>...</sup>` | Superscript | ## CAT tool placeholder tags CAT tools use placeholders/tags that must be preserved exactly: | CAT tool | Examples | | ------------------ | ------------------------------------- | | memoQ | `{1}`, `[2}...{2]`, `{MQ}`, `{tspan}` | | Trados Studio | `<1>`, ``, `<2/>` | | Phrase (Memsource) | `{1}`, `{2}` | ## Tips * Keep tags balanced (for example `text`, not `text`). * If you’re unsure, switch to Tag View and verify the raw tags. * Don’t change tag numbers or names (for example `{1}` → `{2}`), even if the translation “looks fine”. * If you insert a TM match, double-check that tags/placeholders still match the source. Caution For CAT tool workflows, don’t delete or edit placeholder tags unless you know exactly what they represent. # Custom MT endpoint A **Custom MT endpoint** lets you add your own OpenAI-compatible machine-translation service to QuickTrans, alongside the built-in engines (Google, DeepL, Microsoft, …). It is most useful for a **local MT proxy** – a small server that exposes several free MT engines behind a single OpenAI-compatible API – so you can query them all from Workbench without per-engine API keys. It is deliberately separate from the **AI custom endpoint** used for the AI Assistant chat, so you can run an MT proxy for quick lookups *and* point the AI chat at a different custom LLM at the same time. ## When to use it * You run (or have access to) an OpenAI-compatible endpoint that returns translations, e.g. a local MT proxy that maps a `model` name to a specific engine (`google`, `sogou`, `cnpat`, …). * You want fast, free MT in QuickTrans without configuring each engine’s official API key. * You want more than one such endpoint – for example a general proxy and a patent-specific one – each appearing as its own QuickTrans result. ## Set it up 1. Open **Workbench Settings → ⚡ QuickTrans**. 2. Under **MT engines**, tick **Custom MT endpoint (OpenAI-compatible)**. 3. Click **+** next to *Profile* and give the profile a name (e.g. `Local proxy`). 4. Fill in: * **Endpoint URL** – the OpenAI-compatible base URL, e.g. `http://127.0.0.1:1234/v1` * **Model / engine** – the model (or, for a multi-engine proxy, the engine name, e.g. `google`) * **API key** – only if your endpoint requires one; leave blank otherwise * **Show this profile in QuickTrans** – tick to include this profile as a QuickTrans result; untick to keep it configured but hidden 5. Click **💾 Save QuickTrans Settings**. Each profile that is enabled (**Show this profile in QuickTrans** ticked) and has an endpoint appears as its own result in the QuickTrans popup (summoned with **Ctrl+Alt+Q**). Add more profiles with **+** to expose several engines at once; remove one with **−**. Note The “Custom MT endpoint” checkbox is the master on/off for the whole feature; the per-profile “Show this profile in QuickTrans” checkbox lets you pick which of your saved profiles actually appear, so you can keep several configured but show only the ones you want. Note The endpoint must be OpenAI **chat-completions** compatible (it receives a `POST` to `/v1/chat/completions` with `messages` and a `model`, and returns the translation as the assistant message). Workbench sends a strict “translate only” prompt, so the endpoint should return just the translated text. ## Example: a local multi-engine MT proxy A common pattern is a small Python proxy that wraps free web MT engines and presents them as OpenAI “models”. Run it locally (e.g. on `http://127.0.0.1:1234`), then add a Custom MT profile per engine you want, setting **Model / engine** to the engine name the proxy expects. Note Free, unofficial web MT services are best-effort: availability and quality can vary, and they may rate-limit without notice. A proxy keeps that handling outside Workbench. Use each service in line with its own terms, and prefer an official provider API for production work. ## Free Dutch ↔ English engines (via a multi-engine proxy) If your proxy exposes several engines as “models”, set **Model / engine** to the engine’s key. For Dutch ↔ English, these work well: | Model / engine | Notes | | ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `google` | Reliable, fast. | | `microsoft_builtin` | Reliable; good Dutch. | | `modernmt_builtin` | Reliable. | | `lingvanex_builtin` | Works; quality varies. | | `deepl_builtin` | Free DeepL – excellent quality **when available**, but the free endpoint rate-limits aggressively (HTTP 429), so it may intermittently fall back to another engine. | China-focused engines (`sogou`, `transmart`, `niutrans`) and the patent engine `cnpat` are not recommended for Dutch ↔ English. Note These are free, best-effort services and their availability and quality vary; `deepl_builtin` in particular can be throttled. For dependable, production-grade DeepL, use the official DeepL engine (with an API key) built into Workbench’s MT settings. ## Custom MT vs the AI custom endpoint | | Custom MT endpoint | AI custom endpoint | | ------------ | ---------------------------------- | ---------------------------------- | | Lives in | QuickTrans ▸ MT engines | AI Settings ▸ AI/LLM Providers | | Used for | Fast MT results in QuickTrans | AI Assistant chat & AI translation | | Independent? | Yes – configure both at once | Yes | | Profiles | Multiple, each a QuickTrans result | Multiple, one active at a time | # Machine Translation Machine translation is delivered by **QuickTrans** – an always-on-top popup (and dockable panel) with translations from every enabled provider. See [QuickTrans](/workbench/quicktrans/overview/) for the full reference. ## Opening QuickTrans | How | Notes | | ---------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Ctrl+Alt+Q** (⌘⌥Q on macOS) | Opens the QuickTrans always-on-top popup and starts the MT fan-out immediately on the selected text. Auto-copies the current selection so you don’t need a separate Ctrl+C first. | | Editor right-click → ⚡ QuickTrans | Right-click menu in the editor. | | **🔍 Run in SuperLookup** button in the popup header | After you’ve seen the QuickTrans results, click 🔍 to hand the same query off to Workbench’s SuperLookup tab for a richer concordance / termbase / web look-up. | ## Providers QuickTrans supports these MT providers (subject to your API keys and per-provider on/off flags): * DeepL * Google Translate * Microsoft Translator * Amazon Translate * ModernMT * MyMemory (free) Plus optional LLM-based “translation as suggestion” from Claude, OpenAI, Gemini, Mistral, DeepSeek, and a custom OpenAI-compatible endpoint or local Ollama model. ## Configure providers QuickTrans’s provider list and LLM model selectors live in **Workbench Settings → ⚡ QuickTrans**. Click the ⚙ cog icon in the QuickTrans popup header to jump there in one click. Per-provider on/off + LLM model choices persist in `general_settings.json` under `mt_quick_lookup`. ## Language behaviour * QuickTrans uses the active project’s language pair by default. * The popup has its own From / To dropdowns – override per-query without affecting your project settings. ## Performance Provider calls run in parallel (each with a 5 s timeout, overall batch capped at 6 s) so total wall-clock is roughly the slowest single provider, not the sum. Results appear in the popup as they arrive – the first to finish is auto-selected so you can hit Enter without waiting for the slow providers. ## Copying results * Successful results show a **📋 copy button**. * You can also **double-click** a result row to copy the translation. * Number keys **1**–**9** select the corresponding result (1 = first, 2 = second, etc.). Note If a provider call fails, QuickTrans shows the error message in red. Failed providers don’t block the others. # QuickTrans **QuickTrans** shows fast machine translations of the selected text from every enabled provider at once. It runs in two ways: * a **global always-on-top popup**, summoned with **Ctrl+Alt+Q** anywhere on your computer; and * a **docked panel** inside the Workbench grid – it can sit below, above, or to the right of the grid (beside TermLens), showing the same results inline as you move between segments. The rest of this page describes the popup. It’s a single-purpose surface – just translations, no chat – and it stays on top of every other window until you press 1–9 / Enter / click to pick a result, or Esc to dismiss.  ## Opening QuickTrans | Method | Shortcut | Notes | | ---------------------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------ | | Global, from any application | **Ctrl+Alt+Q** (⌘⌥Q on macOS) | Auto-copies the current selection; popup appears with translations | | In-app, from a Workbench grid cell | **Ctrl+Alt+Q** | Same chord – Ctrl+Alt+Q is registered both as a system-wide global hotkey *and* as an in-app QShortcut, so it works wherever you are | | Editor right-click → ⚡ QuickTrans | Right-click menu | Uses the selected text in the cell (or the full cell text if no selection) | After selecting a translation from the global path (Ctrl+Alt+Q from another app), the popup hides itself, returns focus to the source application, and pastes the result over your selection. When invoked in-app, selecting a translation inserts it at the cursor position in the focused grid cell. ## The popup header A row of three controls runs along the top of the popup: * **⚡ Supervertaler QuickTrans** – title * **🔍 Run in SuperLookup** – closes the popup and opens Workbench’s SuperLookup tab with the same query pre-filled and the search auto-fired. Useful when you’ve translated a phrase via QuickTrans and then think “actually, I want to look this up in my TMs / termbases / web resources too” – one click instead of dismissing the popup and pasting the query again * **⚙ Settings** – opens Workbench Settings → ⚡ QuickTrans so you can enable / disable providers and pick LLM models ## Translation results Results arrive as they complete from each provider and are displayed in a numbered list. The first result to arrive is automatically selected, so for the typical “fast provider → press Enter” flow you don’t need to wait for the slow ones. | Method | Action | | ---------------------- | ------------------------------------------- | | **Press 1–9** | Insert the numbered translation immediately | | **Arrow keys + Enter** | Navigate and select | | **Click** | Insert the translation | | **Esc** | Dismiss the popup without inserting | Each translation row shows the provider name on the left and the translated text on the right. ## Supported providers QuickTrans queries up to eleven providers in parallel. Each one is independently enabled / disabled in **Workbench Settings → ⚡ QuickTrans**. **Machine translation engines** (each row needs an API key for that service, except MyMemory): | Engine | API key required? | | -------------------- | ----------------------- | | Google Translate | Yes | | DeepL | Yes | | Microsoft Translator | Yes | | Amazon Translate | Yes | | ModernMT | Yes | | MyMemory | No (free, rate-limited) | **LLM providers** (each row needs an API key for that service; reuses the keys configured in Settings → AI Settings): | Provider | Notes | | -------- | -------------------------------------------------------------------------------------- | | Claude | Pick the model in Settings → ⚡ QuickTrans (e.g. claude-haiku-4-5 vs claude-sonnet-4-6) | | OpenAI | Pick the model (e.g. gpt-5.4-mini vs gpt-5.5) | | Gemini | Pick the model | | Ollama | Local-only; uses the active Ollama model | | Custom | One configurable OpenAI-compatible endpoint (URL + model) | The LLM providers are **disabled by default** – tick them in Settings → ⚡ QuickTrans if you want LLM-based “translation as suggestion” alongside the MT engines. (Ticking all eleven makes for a slow popup; most users keep three or four MT engines plus one LLM.) ## Language pair QuickTrans uses **the active project’s source and target language**. There’s no per-query language override in the popup itself – set the language pair at the project level and QuickTrans inherits it. If no project is open, QuickTrans falls back to English → Dutch (the default for unconfigured installs). ## Configuring providers Open **Workbench → Settings → ⚡ QuickTrans** (or click the ⚙ cog in the popup header) to enable / disable individual providers and pick LLM models. The settings live in `general_settings.json` under the `mt_quick_lookup` key and persist across restarts. Tick a provider, save, then trigger Ctrl+Alt+Q again – the popup picks up the new provider list the next time it opens. ## Tips * **Ctrl+Alt+Q is the fastest way to translate** – select text anywhere, press the shortcut, results appear instantly. The synthetic Ctrl+C happens internally, so you don’t need to copy first. * **Use the 🔍 Run in SuperLookup hand-off** for terminology questions. QuickTrans is great for “how does this phrase translate?”, SuperLookup is great for “have I translated this term before? what does it mean? is it in a termbase?”. * **The popup lives on top of every other window**, so you can summon it from a browser, a PDF reader, your CAT tool, or anywhere – it overlays whatever’s foreground. * **Different from Chat.** QuickTrans gives you N parallel translations from N providers; the Chat tab in Workbench’s right panel is a conversational AI assistant. Use QuickTrans when you want options, Chat when you want a conversation. ## Customising the hotkey The QuickTrans chord can be rebound in **Settings → Keyboard Shortcuts**. The action is called *QuickTrans (instant translation popup)*, default **Ctrl+Alt+Q**. The same chord registers as both an in-app QShortcut and an OS-level global hotkey, so changing it once changes both. ## Related pages * [Machine Translation engines](/workbench/quicktrans/machine-translation/) * [Custom MT endpoint](/workbench/quicktrans/custom-mt-endpoint/) * [SuperLookup Overview](/workbench/superlookup/overview/) * [Keyboard Shortcuts](/workbench/settings/shortcuts/) # Changelog This page shows you where to find **what changed between versions** of Supervertaler. ## ✅ View the full changelog The complete changelog is maintained on GitHub: * [Open CHANGELOG.md on GitHub](https://github.com/Supervertaler/Supervertaler-Workbench/blob/main/CHANGELOG.md) ## What you’ll find there * New features and improvements (what was added) * Bug fixes (what was corrected) * Version numbers and release dates ## Tip If you’re troubleshooting, start by checking whether your issue was already fixed in a newer version. # Contributing Contributions are welcome – bug reports, documentation improvements, and code changes. ## Where to start * Report issues: * Questions / discussion: ## Documentation edits Supervertaler Help is synced from the repository. If you spot missing or unclear documentation, open an issue or submit a pull request. ## Tip When reporting a bug, include: * Your OS * Your Supervertaler version * Steps to reproduce * Any error message text # User Data Folder Supervertaler Workbench keeps your termbases, translation memories, prompt library, settings, and projects in a single user data folder. This folder is **shared with [Supervertaler for Trados](https://docs.supervertaler.com/trados/data-folder/)**, so both programs read and write the same terminology, TMs, and prompts without duplicating files. ## Folder location By default the folder lives in your home directory: ```plaintext Windows: C:\Users\\Supervertaler\ macOS / Linux: ~/Supervertaler/ ``` You can choose a different location during first-run setup. The chosen path is recorded in a small pointer file in your user configuration directory (on Windows, `%APPDATA%\Supervertaler\config.json`), which both programs read so they always agree on where the data lives. ## Folder structure ```plaintext Supervertaler/ │ ├── prompt_library/ Shared │ ├── domain_expertise/ │ ├── project_prompts/ │ └── style_guides/ │ ├── resources/ Shared │ ├── supervertaler.db │ ├── termbases/ │ ├── tms/ │ ├── non_translatables/ │ └── segmentation_rules/ │ ├── workbench/ Supervertaler Workbench only │ ├── settings/ │ │ ├── settings.json │ │ ├── themes.json │ │ ├── shortcuts.json │ │ └── ... │ ├── dictionaries/ │ ├── projects/ │ ├── ai_assistant/ │ ├── voice_scripts/ │ └── web_cache/ │ └── trados/ Supervertaler for Trados only ├── settings/ ├── projects/ └── batch_backups/ ``` ### Shared resources The **prompt library** and **resources** folders are shared between both programs. A prompt you create or edit in Workbench is immediately available in the Trados plugin, and vice versa. The SQLite database (`supervertaler.db`) holds your termbases and translation memories – Workbench has full read-write access to it. ### Program-specific folders Each program stores its own settings, projects, and runtime data in a dedicated subfolder (`workbench/` or `trados/`), so the two never interfere with each other. Workbench’s `workbench/` subfolder holds your `settings/` (including `shortcuts.json` and `themes.json`), custom spellcheck `dictionaries/`, saved `projects/`, AI assistant data, voice scripts, and a web cache. ## Automatic migration If you’re updating from an older version, Workbench reorganises the folder automatically on its next startup. No manual action is required – your settings and data are preserved. ## Related * [Supervertaler for Trados – User Data Folder](https://docs.supervertaler.com/trados/data-folder/) * [General Settings](/workbench/settings/general/) # AutoCorrect while typing Supervertaler can automatically convert straight quotes, three-dot ellipses, and double-hyphen dashes to the correct typographic forms **as you type in the target field**. Quote shapes follow the **target language** – German targets get `„…"`, French targets get `« … »`, Russian targets get `«…»`, English targets get `"…"`, and so on. This feature was requested in [discussion #211](https://github.com/orgs/Supervertaler/discussions/211) and ships from **v1.10.230**. ## Where to find it **Settings → ✍️ AutoCorrect** (its own tab in the Settings sidebar, just below General). The master switch enables or disables all rules at once. Each rule below it can also be toggled individually. This tab **saves automatically** – there is no Save button. Every toggle is written to your settings the moment you click it and takes effect on the **very next keystroke** – no app restart, no grid reload. ## Rules | Rule | Behaviour | Default | | --------------------------------- | --------------------------------------------------------- | ------------------------------------ | | Smart double quotes | `"foo"` → language-correct typographic pair | on | | Smart single quotes / apostrophes | `'foo'` → typographic single pair · `don't` → `don't` | on | | Ellipsis | `...` → `…` | on | | En-dash | `word-- word` → `word– word` | on | | Em-dash | `word--- word` → `word – word` | **off** (the project uses en-dashes) | | French typographic spacing | Insert a narrow non-breaking space before `:` `;` `!` `?` | on for `fr-*` targets | ## Quote shapes by target language The smart-quote rule reads your project’s target language and picks the appropriate shape: | Languages | Open | Close | | -------------------------------------------------------------------------------- | ---------------- | ---------------- | | English, Dutch, Portuguese, Turkish, Romanian, Danish | `"` | `"` | | German, Czech, Slovak, Slovenian, Croatian, Hungarian, Polish *(close uses `"`)* | `„` | `"` / `"` | | French | `« `(with NNBSP) | `»` (with NNBSP) | | Spanish, Italian, Russian, Ukrainian, Norwegian | `«` | `»` | | Swedish, Finnish | `"` | `"` | The engine decides between “open” and “close” shape from what immediately precedes the typed quote – whitespace or an opening bracket → open shape; a letter, digit or closing bracket → close shape. Tag markers (`{1}`, ``, `[2}`) are treated as transparent, so a quote opened straight after an inline tag still gets the opening shape. ## Backspace cancels the last conversion If AutoCorrect converts something you actually wanted to leave alone, press **Backspace immediately**. One Backspace restores your literal typing (the straight quote, the three dots, the double hyphen, etc.) – exactly as it works in Word and memoQ. The next keystroke clears that one-shot undo memory, so the safety net is only available for the conversion you just made. ## What AutoCorrect does *not* touch * **Inside tag markers** (`{1}`, `[2}`, ``). Auto-correcting inside a tag would corrupt the boundaries and break the round-trip back to your source format, so the engine skips these. * **Paste**. Pasting a block of text never triggers any rule – only single typed characters do. If you want the engine to clean up pasted text, do it explicitly with Find & Replace. * **Programmatic content** (loaded translations, MT/TM insertions, Copy Source → Target). Same reason – these aren’t user keystrokes. * **Dictation**. Voice-typed content arrives via a different input path and is not auto-corrected. * **The source column**. AutoCorrect is target-only. ## Turning a single rule off temporarily Use the per-rule toggles on the **Settings → ✍️ AutoCorrect** tab. Because the tab saves automatically, both the master switch and the per-rule checkboxes are honoured on the next keystroke without clicking anything else. There’s no per-segment override yet – if you want one in a future version, please open an issue. ## See also * [Settings → General](/workbench/settings/general/) * Tracking issue: [#213 – Typographic auto-convert / AutoCorrect-while-typing system](https://github.com/Supervertaler/Supervertaler-Workbench/issues/213) * Original request: [Discussion #211](https://github.com/orgs/Supervertaler/discussions/211) # Backup Supervertaler protects your work with two independent backup mechanisms, both on the **Settings → 💾 Backup** tab. Use the **?** button on that tab (or press **F1**) to return to this page. ## Auto Backup (time-based) Automatically saves the current project at a regular interval to guard against crashes and forgotten saves. * **Enable automatic backups** – turn the timer on or off. * **Backup interval** – how often to save, in minutes (default **5**, range 1–60). Each run saves the project file and exports a `_backup.tmx` alongside it. This **overwrites** the working files in place – it keeps the latest state current, but it is not a history you can step back through. For that, use timestamped backups below. ## Timestamped project backups (every N saves) Keeps **immutable, dated snapshots** of the project file (`.svproj`) so you can roll back to an earlier state – like a lightweight version history. * **Keep timestamped project backups** – enable or disable the feature. * **Back up every N saves** – how often a snapshot is taken, counted in save operations (default **1** = every save). Both manual saves (Ctrl+S) and the timed auto-backup above count toward N. * **Keep the last K backups** – how many snapshots to retain per project (default **100**). Older ones are pruned automatically. Snapshots are written to a dedicated folder under your user-data location: ```plaintext \workbench\backups\\_YYYYMMDD-HHMMSS.svproj ``` Use the **Open folder…** button on the Backup tab to jump straight there. Note Taking a snapshot just copies the project file you already saved, so it adds no noticeable delay – backing up on every save is fine even on large projects. ### Restoring a backup 1. Click **Open folder…** on the Backup tab (or browse to the path above). 2. Find the snapshot with the timestamp you want (filenames sort chronologically). 3. Copy it somewhere safe and rename it (e.g. drop the timestamp), then open it from **Project → Open**, or replace your current `.svproj` with it while Supervertaler is closed. Tip Because every save can be a snapshot and old ones are pruned for you, if something ever goes wrong you can almost always step back to a known-good version from a minute or two earlier. ## Related * [General Settings](/workbench/settings/general/) # Fonts Font settings control the typeface and size used in the translation grid and the companion tabs. ## What you can change * **Font family** – any font installed on your system; the dropdown shows available fonts * **Font size** – point size for the grid; companion tabs use the same family at their own size * **Global UI font scale** – a single slider that scales every UI element (menus, tabs, settings, Chat panel, Clipboard history, SuperLookup, status bar) at once ## Choosing a font * For general translation work, a clear humanist sans-serif (Segoe UI, Inter, Calibri) keeps long sessions comfortable * For technical or code translation, a monospaced font (Consolas, JetBrains Mono) can help align numbers and symbols * For right-to-left languages (Arabic, Hebrew), choose a font with good RTL glyph coverage ## Global UI font scale (Retina / high-DPI displays) Settings → AI Settings → **🖥️ Global UI Font Scale** holds a single slider (50%–200%, default 100%) that scales the entire application UI – not just the grid. Useful when you find Qt’s defaults uncomfortably small on a MacBook Retina screen, a 4K monitor, or any high-DPI display. The slider covers: * The grid (segment numbers, type column, source and target text) * Companion tabs (Clipboard 3-column tree, SuperLookup web resources, Voice command table, Chat panel) * QuickTrans always-on-top popup * Tabs, settings panels, AI tools, status bar, menus * Termbase and TM panes Apply the change with the **Apply** button next to the slider; most areas update immediately. Lazy-constructed widgets (Clipboard, SuperLookup, Voice) pick up the new size when they are next opened, so switch away from a companion tab and back once after changing the slider. If you’ve also customised the grid font size (above), that value still applies on top of the scale – so a 12 pt grid font at 150% renders at 18 pt. Grid zoom (Ctrl+= / Ctrl+-) continues to work at any scale. ## Tips * Font changes apply immediately in the grid – no restart needed * If glyphs for a specific language appear as boxes, install a font with full Unicode coverage for that script (Noto Sans is a good all-rounder) * On a 4K or Retina display, try 125% or 150% UI scale before reaching for individual font-size sliders – it keeps every panel proportional * The QuickTrans popup’s header controls (🔍 Run in SuperLookup, ⚙ Settings) deliberately don’t scale with the slider, because they live in fixed-size buttons and scaling the glyph alone would overflow them ## Related pages * [View Settings](/workbench/settings/view/) * [Theme (Light/Dark Mode)](/workbench/settings/theme/) # General ## Where to find settings Open **Settings** from the main toolbar or the **View** menu. Settings are organised into tabs across the top of the settings panel. ## AI Settings * **Provider** – select OpenAI, Anthropic (Claude), Google Gemini, Mistral, or a custom OpenAI-compatible endpoint * **API key** – enter and save your key for the selected provider; keys are stored locally in your user data folder * **Model** – choose which model to use for AI translation and the Chat assistant * **Temperature** – controls how creative vs. literal the AI output is (lower = more consistent) * **Max tokens** – upper limit on response length See [Setting Up API Keys](/workbench/get-started/api-keys/) for step-by-step instructions. ## Project settings * **Default source language / target language** – pre-filled when creating new projects * **Autosave interval** – how often the current segment is saved automatically (in seconds); set to 0 to disable ## Voice settings The [🎤 Voice top tab](/workbench/voice/overview/) contains all voice command and dictation settings (engine, model, sensitivity, push-to-talk mode). They are not duplicated here – open the Voice tab directly to configure them. ## Related pages * [Setting Up API Keys](/workbench/get-started/api-keys/) * [AI Translation Overview](/workbench/ai-translation/overview/) * [Theme (Light/Dark Mode)](/workbench/settings/theme/) * [Keyboard Shortcuts](/workbench/settings/shortcuts/) # Language (UI Translation) Supervertaler Workbench can display its menus and settings labels in languages other than English. Translations are contributed by the community as **XLIFF 1.2** files – the industry-standard interchange format that every CAT tool reads natively. This page covers: * How to pick your display language * Which languages are currently available * How the translation files work (file location, format) * How to contribute a translation Note **Current scope (MVP, v1.10.208).** This first translation pass covers the **menu bar**, **Settings tab labels**, and **General-tab group titles** – around 180 strings. Dialog bodies, error messages, status-bar text, and per-cell tooltips remain in English for now. Subsequent passes will widen coverage as translators contribute. ## Picking your display language 1. Open **Settings → General → 🌐 Language**. 2. Use the **Display language** dropdown to choose a locale. 3. Click **OK** to close Settings. 4. **Restart Supervertaler** for the new language to take effect.  Settings → General → Language dropdown  The Workbench interface after picking a locale – here localised to Dutch ### Locale options | Entry | Meaning | | -------------------------- | ------------------------------------------------------------------------------------------------------------------------- | | **System default** | Use whatever language your operating system reports. Falls back to English if no translation file exists for that locale. | | **English** | Source language – no translation file needed. | | Locales with a `.xlf` file | Use that translation. | | `[no translation yet]` | Locale code is recognised but no `.xlf` has been contributed yet. Picking it falls back to English. | If a locale is partially translated, the strings that *have* been translated are used; the rest fall through to English. Partial coverage is fine. ### Why a restart is required Most Supervertaler dialogs are hand-coded rather than built with Qt Designer, which means they don’t automatically refresh their text when the language changes mid-session. A restart is the simpler approach for v1; live language switching may come in a future version. ## Where the translation files live Each locale’s translation lives in a single `.xlf` file in the **`translations/`** folder: * **Installed Windows build:** alongside `Supervertaler.exe` (open the install folder, find `translations\`) * **From source:** in the repository root, `translations/` * **macOS:** inside the `.app` bundle at `Supervertaler.app/Contents/Resources/translations/` File names follow the pattern `supervertaler_.xlf`: ```plaintext translations/ ├── supervertaler_template.xlf <- source-only English template (auto-generated) ├── supervertaler_zh_CN.xlf <- Simplified Chinese ├── supervertaler_zh_TW.xlf <- Traditional Chinese ├── supervertaler_pl.xlf <- Polish └── ... ``` **You can drop a new locale’s `.xlf` into this folder manually** – the next launch will pick it up and the Language dropdown will offer it. No reinstall, no rebuild. ## What’s inside an XLIFF file? The format is standard **XLIFF 1.2** – the same format Workbench imports through **Project → Import → Import Document…** (using the XLIFF filter). Every CAT tool in regular use reads it natively. A `.xlf` file has one entry per translatable string. Each entry looks like this: ```xml &Project 项目(&P) SupervertalerQt Supervertaler.py 9698 ``` What each piece means: * **``** – the original English string. Never edit this. * **``** – your translation. Set the `state` attribute to `translated` (or `signed-off` / `final`) when you’re happy with it. Targets left as `state="needs-translation"` are skipped at runtime – Workbench shows the English source instead. * **``** – where the string appears in the UI. The `x-qt-context` value tells you the dialog/class; the `sourcefile` and `linenumber` are pointers into the source code (rarely needed for translation, but handy for troubleshooting). * **`&`** – XML-escaped `&`. The character marks the **next letter as a keyboard mnemonic** – `&Edit` becomes underlined **E**dit, activated by **Alt+E**. Translations should preserve mnemonics, often by putting them in parentheses (`编辑(&E)` for Chinese). ## How to contribute a translation The whole flow is designed so you can use whichever CAT tool you already work in. No new tooling required. ### Quick version 1. Grab `translations/supervertaler_template.xlf` from the [GitHub repo](https://github.com/Supervertaler/Supervertaler-Workbench/tree/main/translations). 2. Save a copy as `supervertaler_.xlf` (for example `supervertaler_de.xlf` for German). 3. Open it in your CAT tool (Trados, memoQ, Phrase, OmegaT, or Workbench itself). 4. Set the target language on the `` element (`target-language="de"` for German). Most CAT tools do this for you when they ask which target language you’re translating into. 5. Translate the strings. Mark each as **confirmed/translated/approved** in your tool – this sets the `state="translated"` attribute behind the scenes. 6. Save / export back to XLIFF. 7. Open a PR on the [Supervertaler-Workbench repo](https://github.com/Supervertaler/Supervertaler-Workbench/pulls) adding your `.xlf` file under `translations/`. ### CAT-tool specifics **Workbench itself:** Project → Import → **Import Document…** and choose the generic XLIFF filter (not SDLXLIFF or MQXLIFF). Translate in the editor, then export back via Project → Export → **Export Translated Document…**. **Trados Studio:** File → Open → Translate Single Document → select the `.xlf`. Studio recognises XLIFF 1.2 out of the box. Save Target As to export. **memoQ:** Project → Import documents → select the `.xlf`. Use the standard XLIFF filter. **Phrase TMS:** Upload as a regular bilingual file. **OmegaT:** Drop into the project’s `source/` folder. OmegaT outputs the completed file to `target/`. **Poedit:** Although Poedit is best-known for `.po` files, recent versions also handle `.xlf` natively. ### Notes for translators * **Mnemonics (`&letter`)** mark keyboard accelerators. Preserve them in the translation. Place them where the underlined letter feels natural in your language. For Chinese / Japanese / Korean, the convention is to put the mnemonic in parentheses after the term: `编辑(&E)`. * **Avoid mnemonic collisions** within the same menu. Two items both using `&S` will cause one to silently lose the shortcut. * **Emoji** (📁 🔍 ⚙️ etc.) stay in the translation – they’re part of the visual identity. * **Newlines (`\n`)** in source strings should be preserved. * **Placeholders** like `{0}` or `%1` are not in this MVP’s strings, but if you see one, leave it verbatim. ## How translations are picked up Each launch, Supervertaler: 1. Reads `general.ui_locale` from the unified `settings.json`. 2. Resolves `"system"` to the operating system’s locale via Qt’s `QLocale.system()`. 3. Looks for `translations/supervertaler_.xlf` next to the executable. 4. Parses the XLIFF, collects all `` entries whose `` is in a “done” state (`translated`, `signed-off`, `final`, or stateless). 5. Installs a `QTranslator` on the QApplication so every `tr()` call resolves through that dictionary. 6. If the file is missing, the locale has zero finished translations, or the locale is `en` / unknown – English is used silently. Total cost at startup: about 10 ms. Negligible against the rest of the cold-start. ## Troubleshooting **The dropdown shows my locale as `[no translation yet]` even though I added the file.** Make sure the file: * Lives in `translations/` (next to `Supervertaler.exe`, or in the source-tree root) * Is named `supervertaler_.xlf` exactly – matching the locale code in the dropdown (case-sensitive) * Is valid XML (a single missing `` will fail the load silently) **The dropdown picks the right locale, but the menu still shows English.** Check that: * Each `` element has `state="translated"` (or `signed-off` / `final`), not `state="needs-translation"` * The `` actually contains text, not just whitespace * The `` text matches the codebase exactly (case, punctuation, spaces, emoji) * You’ve actually restarted Supervertaler **XML parse errors at startup.** Open the `.xlf` in a text editor. Common causes: * Unescaped `&` in a translation – use `&` instead * Mismatched `<` / `>` in a translation – use `<` and `>` * A CAT tool reordered elements in a way Workbench doesn’t expect Open an issue with the `i18n` label on the [Workbench tracker](https://github.com/Supervertaler/Supervertaler-Workbench/issues) and attach the failing `.xlf`. ## See also * [`translations/TRANSLATING.md`](https://github.com/Supervertaler/Supervertaler-Workbench/blob/main/translations/TRANSLATING.md) – the in-repo contributor guide with extra detail * [General Settings](/workbench/settings/general/) – the parent Settings tab * Tracker issues [#178](https://github.com/Supervertaler/Supervertaler-Workbench/issues/178) and [#190](https://github.com/Supervertaler/Supervertaler-Workbench/issues/190) – the original i18n requests # Customising Shortcuts Supervertaler has one keyboard shortcut per action. The same combination works whether Supervertaler is the focused application or whether you trigger it from another app – there’s no longer a separate “Global” entry to keep in sync with the in-app one. ## Managing shortcuts Open **Settings → Keyboard Shortcuts**. Each row is one action. Click a row to edit, press the new combination, hit OK. Changes apply immediately – no restart needed. Rows whose action label starts with 🌍 also register as an OS-level global hotkey, so they fire from any application. The rest are in-app only (e.g. segment navigation, match insertion). ## Default shortcuts that work everywhere The 🌍 actions and their out-of-the-box bindings: | Action | Default | Notes | | ------------------------------ | -------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | Open Clipboard | **Ctrl+Alt+C** (Win/Linux) / **⌘⌥C** (macOS) | Auto-copies the current selection, then opens Workbench’s Clipboard tab | | Open SuperLookup | **Ctrl+Alt+L** / **⌘⌥L** | Auto-copies the current selection, then opens Workbench’s SuperLookup tab with the text pre-filled and the search auto-fired | | QuickTrans | **Ctrl+Alt+Q** / **⌘⌥Q** | Instant translation popup; auto-copies the selection | | Voice dictation / push-to-talk | **Ctrl+Shift+Space** / **⌘⇧Space** | Toggles recording; a ”🎤 Listening…” toast confirms the mic is live | | Voice commands push-to-talk | **Ctrl+Alt+V** / **⌘⌥V** | Hold to listen for voice *commands* only – for pairing Workbench’s command listener with an external dictation app | | Voice Always-On (toggle) | **Ctrl+Alt+O** / **⌘⌥O** | Continuous listening on/off | Note **Ctrl+Alt+K** used to summon a floating Supervertaler Sidekick window through v1.10.3. That window was retired in v1.10.4 and the chord is now unbound by default. The Clipboard Manager, Voice, and SuperLookup tabs are reachable via the dedicated hotkeys above; Chat lives in the AI tab and in Workbench’s right panel. Rebind any of these in **Settings → Keyboard Shortcuts** by clicking the row and pressing a new combination. ### Running Workbench alongside Supervertaler for Trados The two products share a number of shortcuts on purpose – `Alt+↓` adds a term in both, `Alt+1`…`Alt+9` inserts term *n* in both, `Ctrl+Alt+T` opens the term dialogue in both. That is safe, because those are ordinary in-app shortcuts: whichever window you are working in is the one that responds. The **🌍 global hotkeys in the table above are the exception**. They are registered with the operating system, so they fire no matter which application is in front – including while you are in Trados. That is the point of them (you can grab a selection from any app), but it means a global hotkey must not be a key Trados also uses, or one press would do two things at once. None of them currently clash. If you rebind one, keep it clear of the [Supervertaler for Trados shortcuts](/trados/keyboard-shortcuts/) – and note the symptom to watch for: a *hold*-style hotkey pressed together with a *toggle*-style one leaves the toggle switched on after you let go, with nothing visible having started it. ## macOS vs Windows: symbols and modifier names The two platforms use different conventions for naming and drawing modifier keys. Supervertaler shows the platform-native symbols in the UI, but it’s useful to know what each one means: | Symbol | macOS name | Windows / Linux name | Physical key | | ------ | -------------- | -------------------- | ------------------------------------------ | | **⌘** | Command (Cmd) | – | The key with the Apple/Command glyph | | **⌃** | Control (Ctrl) | Ctrl | The Control key | | **⌥** | Option (Opt) | Alt | The Alt key (top of the Option key on Mac) | | **⇧** | Shift | Shift | The Shift key | So a shortcut shown as **⌃⌘L** on macOS is read “Control + Command + L”. See the next section for how Supervertaler stores that internally and why the stored name doesn’t always match the Mac symbol. ## What “Ctrl” means inside Supervertaler Internally, shortcuts are stored in Qt’s cross-platform format, which uses **Ctrl**, **Alt**, **Shift**, and **Meta** as labels. Qt swaps **Ctrl** and **Meta** on macOS so that the same shortcut string works on every platform. The mapping: | Stored as… | …means on Windows / Linux | …means on macOS | | ---------- | ------------------------- | --------------- | | `Ctrl` | Ctrl | **Cmd** (⌘) | | `Alt` | Alt | **Option** (⌥) | | `Shift` | Shift | **Shift** (⇧) | | `Meta` | Windows key | **Control** (⌃) | You only ever see this if you export shortcuts to JSON or look at the cheatsheet HTML – the UI itself always shows you platform-native symbols on macOS and plain names on Windows. The default for SuperLookup is therefore stored as `Ctrl+Alt+L` and displayed as **Ctrl+Alt+L** on Windows and **⌘⌥L** on macOS, both of which fire the same physical chord on each platform. ## Per-platform notes **macOS** Global hotkeys require Accessibility permission on whichever binary launched Python: * Bundled `Supervertaler.app` → add **Supervertaler** in System Settings → Privacy & Security → Accessibility * Launched from `Terminal.app` → add **Terminal.app** instead * Launched from iTerm2 → add **iTerm2.app** instead Also requires the `pyobjc-framework-Cocoa` Python package (`pip install pyobjc-framework-Cocoa`); the bundled `.app` ships with it. The Status indicator on the right-hand side of Settings → Keyboard Shortcuts shows **Active (via NSEvent)** when global hotkeys are working on macOS. **Windows** Global hotkeys are registered via the native `RegisterHotKey` API, which consumes the keystroke at the OS level. The combination is reserved for Supervertaler whenever it’s running. If another app has already claimed the same combination, Supervertaler logs a `failed_hotkeys` warning and that one combination won’t fire – re-bind to something free in Settings → Keyboard Shortcuts. **Linux** Global hotkeys go through `pynput`, which uses XGrabKey under X11. If hotkeys silently don’t fire, your user may need to be in the `input` group (`sudo usermod -aG input $USER`, then log out and back in). ## Quick-lookup tab keyboard navigation When you’ve summoned the Clipboard, SuperLookup, or Voice tab via a global hotkey, these shortcuts work straight away – no clicking around to land your focus first: | Shortcut | Action | | --------- | ---------------------------------------------------------------------------------------------------------- | | **Esc** | Hide Workbench to the system tray (quick-lookup tabs only – Editor / Settings / etc. keep the natural Esc) | | **↑ / ↓** | Navigate within the focused column (e.g. clipboard text history, snippet list) | | **← / →** | Move focus between columns in the Clipboard tab (Text → Images → Menu) | | **Enter** | Activate the selected item (paste clip, run snippet, fire conversion) | ### Pressing Esc dismisses Workbench to the tray On the surfaces you summon with a global hotkey – SuperLookup, Clipboard, and Voice – pressing **Esc** hides Workbench back to the system tray. Handy when you’re using Workbench as a popup utility from another app: hotkey to summon, Esc to dismiss. * **On SuperLookup**: Esc unconditionally hides Workbench, even when the cursor is in the search box. SuperLookup is mostly a one-shot query, so there’s nothing worth keeping if you change your mind. * **On Clipboard and Voice**: Esc hides Workbench *unless* the focused widget is a text input (search field, command editor, etc.) – in those cases Esc behaves the way it does in any other app (clears the field, closes a dropdown, etc.). * **On Editor, TMs, Termbases, AI, Settings**: Esc keeps its natural editor / dialog / combo-box behaviour. Workbench is never hidden by accident from the surfaces where you actually do work. ### Tray quick-jump menu Right-click the Workbench tray icon (the orange **Sv**) for a menu with **Show Workbench**, **Open SuperLookup**, **Open Clipboard**, **Open Voice**, **Open Settings**, plus toggles for **Close to tray** and **Start with computer**. ## Editor shortcuts The editor (translation grid) has its own set of shortcuts for navigation, match insertion, term operations, and so on. See [Editor Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) for the full list. ## Exporting a printable cheatsheet The settings page has an **Export Cheatsheet (HTML)** button on the right-hand panel. It writes a self-contained HTML file showing every shortcut grouped by category, with the platform-native symbols already substituted in. Print it or save it as PDF. ## Related pages * [Editor Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) * [Voice Commands & Dictation](/workbench/voice/overview/) * [Clipboard Manager](/workbench/clipboard/overview/) # Termbase Settings The **Termbase settings** box (Settings → General) controls how termbase matches are displayed while you translate. ## Highlight termbase matches in source cells When enabled, termbase matches are highlighted with coloured backgrounds directly in the source column of the grid. Higher-priority terms are shown in darker blue, lower-priority terms in lighter blue – similar to memoQ’s termbase highlighting. This gives you instant visual feedback about which words in the source are covered by your termbases. ## Hide shorter termbase matches included in longer ones When enabled, shorter terms that are fully contained within a longer matched term are hidden from the results. For example, if both *cooling* and *cooling system* match, only *cooling system* is shown. This reduces clutter in the translation results panel when overlapping terms are present. # Theme Supervertaler supports light and dark themes. Switch between them in **Settings → Theme** or via the toolbar theme toggle. ## Why use Dark Mode * Better comfort in low-light environments * Reduced eye strain during long evening sessions * Lower screen brightness without sacrificing readability ## Switching themes The change takes effect immediately – no restart needed. All panels (grid, companion tabs, dialogs, popups) switch at once. ## Tips * If any UI element looks visually wrong after switching (rare Qt repaint quirk), try switching to a different tab and back, or restarting the app * Dark mode does not affect PDF Rescue’s OCR output or exported DOCX files – those are document colours, not UI colours ## Related pages * [View Settings](/workbench/settings/view/) * [Font Customisation](/workbench/settings/fonts/) # TM Settings The **TM settings** box (Settings → General) controls how translation memory matches are inserted and propagated as you work. ## Auto-fill empty segments with 100% TM matches When you select an empty segment that has a 100% TM match, its translation is filled in automatically. This saves time on repetitive content – exact matches appear without any manual action. Note Only empty segments are auto-filled. Segments that already contain a translation are left untouched. * **↳ Mark auto-filled segments as confirmed** – when enabled, an auto-filled segment lands as **confirmed**. Otherwise the auto-filled translation is inserted as a **draft** for you to review. ## Auto-propagate confirmed translations to identical segments When you confirm a segment, its translation is copied to every other segment whose source text is identical. This is the memoQ/Trados-style propagate-on-confirm behaviour – translate a repeated sentence once, confirm it, and the rest of the document catches up automatically. By default, only **empty** identical segments are filled, and they are inserted as drafts. * **↳ Mark propagated segments as confirmed** – propagated segments land as **confirmed** instead of **draft**. * **↳ Overwrite existing translations when propagating** – propagation also replaces existing target content in identical segments. When disabled, only empty segments are filled. ## Auto-confirm 100% TM matches when navigating (Ctrl+Enter) When enabled, pressing **Ctrl+Enter** automatically inserts, confirms, and skips past any segment that has a 100% TM match, so you can move quickly through perfect matches. * **↳ Also overwrite existing translations with 100% TM matches** – auto-confirm also replaces existing target content (including pre-translations or machine translations) with the 100% match. ## TM Save Mode Controls what happens when the same source segment is saved to the TM more than once. * **Save all translations (with timestamps)** – keeps every version of a translation for a given source, with timestamps. The most recent translation is preferred when showing matches. * **Save only latest translation (overwrite)** – keeps only the most recent translation, overwriting older ones. This prevents the TM from growing with obsolete entries. *(Default.)* # View View settings control how the translation grid and the side panels look while you translate. ## Grid display * **Show invisibles** – reveals spaces, tabs, and line breaks as visible markers; useful for catching trailing whitespace * **Tag colour** – the highlight colour used for inline formatting tags (``, ``, etc.) * **Row height** – compact, normal, or spacious; affects how many segments you see at once without scrolling ## Right panel (Chat, matches) * **Right panel default width** – default width of the right-hand panel that hosts the Match Panel and the 💬 Chat tab. Can also be dragged at runtime; the width is remembered across sessions. ## Tips * If tags are hard to see, increase tag colour saturation or switch to **Tag View** (shows placeholder boxes instead of raw tag text). * For long sessions, try a slightly larger row height – it reduces eye strain when scanning for specific segments. * Dark mode is set in [Theme](/workbench/settings/theme/), not here. ## Related pages * [Theme (Light/Dark Mode)](/workbench/settings/theme/) * [Font Customisation](/workbench/settings/fonts/) * [Keyboard Shortcuts](/workbench/settings/shortcuts/) # Overview SuperLookup is your unified concordance and research hub, bringing together all lookup resources in one place. If you’ve used **LogiTerm Pro**, the idea will feel familiar: one search box across your termbases, translation memories, and web resources. It’s a top tab in Workbench (🔍 SuperLookup), alongside Editor, TMs, Termbases, Clipboard, Voice, and Settings. ## Opening SuperLookup | How | Shortcut | Notes | | ---------------------------------- | ----------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | From the translation grid | **Ctrl+K** | Selects the 🔍 SuperLookup top tab; selected text is used as the search query automatically | | From any application (system-wide) | **Ctrl+Alt+L** | Select any text in any app, press the shortcut, and Workbench opens with the SuperLookup tab forward, text pre-filled, and the search auto-fired. The sub-tab it lands on is configurable – see **Configurable landing tab** below. | | From any application (system-wide) | **Ctrl+Alt+Q** | Opens the [QuickTrans always-on-top popup](/workbench/quicktrans/overview/) with parallel translations from every enabled provider. Different from SuperLookup – use SuperLookup for terminology lookup, QuickTrans for fast MT options. | | Via the system tray | Right-click the orange Sv icon → **Open SuperLookup** | Useful when the global hotkey is taken by another app | ## Configurable landing tab By default, Ctrl+Alt+L lands on the **Termbases** sub-tab. You can change this in **SuperLookup Settings → “Ctrl+Alt+L lands on”** – pick from QuickTrans, TMs, Termbases, or Web Resources. The choice persists across restarts. Whichever sub-tab you choose is fired immediately on the hotkey; the others are deferred until you actually navigate to them, so you don’t pay the cost of work you may not look at. ## Tabs ### TM Matches Search your Translation Memories for similar text: * Fuzzy matching with percentage scores * Horizontal (table) or vertical (list) view toggle * Source TM column shows which TM the match came from * Search direction: Both, Source only, or Target only ### Termbase Matches Search your termbases: * Shows Source, Target, Domain, Notes columns * Right-click to “Edit in Termbase” * Direction and language filters (Both / Source / Target + From/To) ### Machine Translation Get instant MT from multiple providers: * Google Translate, DeepL, Microsoft Translator * Amazon Translate, MyMemory, ModernMT Configure providers in **Settings → MT Settings**. ### Web Resources Quick access to online reference sites: IATE, Linguee, ProZ.com, Reverso Context, Wikipedia, Wiktionary, Google, Google Patents, Juremy, AcronymFinder, BabelNet, and more. Web resource tabs maintain login sessions between searches, so you stay logged in to sites like ProZ.com. ## Search controls **Language filters** – use the **From** and **To** dropdowns to filter by language pair. Auto-populated from your TMs and termbases. **Search direction** – **Both** searches source and target columns; **Source** and **Target** restrict to one side. **Search box** – type a query and press Enter (or click 🔍). When Ctrl+K or Ctrl+Alt+L opens SuperLookup, the selected text is placed in the search box and the search runs immediately. ## Tips * Double-click a result to copy it to the clipboard. * Right-click a result for additional options. * Press **Esc** to hide Workbench back to the system tray once you’ve grabbed what you needed – Esc unconditionally dismisses from SuperLookup, regardless of whether you were typing in the search box. *** ## Related pages * [QuickTrans](/workbench/quicktrans/overview/) * [TM Concordance Search](/workbench/superlookup/tm-search/) * [Termbase Search](/workbench/superlookup/termbase-search/) * [Web Resources](/workbench/superlookup/web-resources/) # Termbase Search SuperLookup’s **Termbases** tab searches your Supervertaler termbases for preferred terminology. ## Open it * **In Supervertaler:** press `Ctrl+K` (or open the **SuperLookup** tab, or **Tools → SuperLookup**). * **From any application:** press `Ctrl+Alt+L` (system-wide hotkey). ## How to search 1. Type (or paste) a term into the search box. 2. Press `Enter` (or click **Search**). 3. Optional filters: * **Direction:** Both / Source / Target * **From / To:** filter by termbase language pair (leave as **Any** to search across languages) ## What you’ll see Results are shown in a table with: * **Source** * **Target** * **Termbase** (termbase name) * **Domain** * **Notes** The current search term is highlighted in the results. ## Actions * **Double-click** a row to copy the translation. * **Copy Translation** copies the selected target term. * **Add to Termbase** opens a dialog to add a new term pair (you’ll be prompted to pick a writable termbase). ## Edit in Termbase (jump to the entry) Right-click a result to open a context menu with: * **Edit in Termbase: …** This navigates to the **Termbases** tab, selects the termbase, and filters the terms list to the source term. ## Termbase selection In **SuperLookup → Settings → Termbases**, you can pick which termbases SuperLookup searches. Note If you don’t select any termbases in the SuperLookup Settings tab, SuperLookup searches **all available** termbases. # TM Search (Concordance) SuperLookup’s **TMs** tab lets you do fast concordance searches in your Translation Memories (“find where I translated this before”). ## Open it * **In Supervertaler:** press `Ctrl+K` (opens SuperLookup with the current selection, if any). * **From any application:** press `Ctrl+Alt+L` – a true system-wide hotkey (registered natively on Windows; no AutoHotkey required). ## How to search 1. Type (or paste) text into the search box. 2. Press `Enter` (or click **Search**). 3. Optional filters: * **Direction:** Both / Source / Target * **From / To:** language pair filter (leave as **Any** to use all languages) ## Views At the top of the tab you can switch between: * **Horizontal (Table):** match % + Source/Target side-by-side. * **Vertical (List):** stacked Source/Target entries (classic concordance layout). ## Actions * **Double-click** a result to copy the target text. * **Copy Target** copies the selected target. * **Insert Target** copies the target and prompts you to paste with `Ctrl+V` in your active application. ## TM selection In **SuperLookup → Settings → Translation Memories**, you can choose which TMs to search. Note If you don’t select any TMs in the SuperLookup Settings tab, SuperLookup searches **all available** TMs. # Web Resources SuperLookup’s **Web Resources** tab gives you a one-click sidebar of reference sites for terminology and research. ## How it works * You use the **main SuperLookup search box** (top of the window) and click **Search**. * The Web Resources tab uses your **From → To** language direction when building URLs. ## Browser modes In the left sidebar you can choose a mode: * **Embedded:** opens sites inside Supervertaler (requires `QtWebEngine`). * Uses a persistent browser profile so logins/cookies are kept between sessions. * **External:** opens your default browser. Note When Embedded mode is available, SuperLookup can pre-load searches for all resources at once. ## Search options * Select a single resource (e.g. IATE) and click **Search**. * Click **Search All** to load results for all resources (Embedded mode). * Use **Open in Browser** to open the last search URL in your default browser. ## Included resources The sidebar includes (by default): * IATE * Linguee * ProZ.com * Reverso Context * Google Search * Google Patents * Wikipedia (Source) * Wikipedia (Target) * Juremy * beijer.uk * AcronymFinder * BabelNet * Wiktionary (Source) * Wiktionary (Target) ## Show/hide resources In **SuperLookup → Settings → Web Resources**, you can toggle which sites appear in the sidebar. # Sending Terms to the AI Supervertaler can send your termbase entries to the AI as reference during translation, so the model uses your approved terminology instead of guessing. This is on a per-termbase basis and takes one click to enable. ## How to enable it Each termbase in the **Termbase Manager** has an **AI** checkbox (the orange/purple tick), separate from the Read/Write activation checkboxes. 1. Open the **Termbases** tab. 2. Find the termbase whose terms you want the AI to use. 3. Tick its **AI** checkbox. That’s the only setup step. From then on, matching terms are automatically included in every translation prompt for the current project. Note The **AI** checkbox is independent per termbase. You might, for example, feed a small client-approved glossary to the AI while keeping a large general reference termbase for lookups only. ## What happens during translation When you translate a segment, Supervertaler: 1. Scans the source text of that segment. 2. Finds any terms from your AI-enabled termbases that actually appear in it. 3. Adds them to the prompt sent to the AI under a **TERMBASE** heading, instructing the model to use those approved terms. The section added to the prompt looks like this: ```plaintext # TERMBASE Use these approved terms in your translation: - machine learning → machinaal leren - click → klikken - widget → ⚠️ DO NOT USE: widget ``` ### Only relevant terms are sent Supervertaler sends **only the terms that appear in the current segment**, not your entire termbase. Matching is whole-word for spaced languages and substring-based for CJK/Thai. This keeps each prompt focused, avoids diluting the AI with irrelevant terminology, and saves tokens. ### Forbidden terms If you mark a term as **forbidden** in the termbase, the AI is explicitly told **DO NOT USE** that translation. This is useful for steering the model away from a wrong-but-tempting rendering, or away from an old term a client has since replaced. ## Tips * **Keep AI-enabled termbases focused.** Very large termbases trigger a warning, because sending a lot of terminology can dilute translation quality. Thanks to per-segment filtering this rarely bites in practice, but a tight, curated glossary gives the best results. * **Works everywhere.** Term injection applies to single-segment translation, batch translation, and keyboard-shortcut translation alike. * **Combine with TM.** Fuzzy TM matches are injected alongside termbase terms, so the AI gets both your approved terminology and your existing translations as reference. ## See Also * [Termbase Basics](/workbench/termbases/basics/) * [Importing Terms](/workbench/termbases/importing/) * [AI Translation Overview](/workbench/ai-translation/overview/) * [Prompts](/workbench/ai-translation/prompts/) # Termbase Basics Termbases help ensure consistent terminology across your translations. ## What is a Termbase? A termbase is a database of terms with their translations: | Source (EN) | Target (NL) | Domain | Notes | | ---------------- | --------------- | ------ | --------------- | | software | software | IT | Don’t translate | | click | klikken | IT | Verb | | machine learning | machinaal leren | AI | Official term | ## Why Use Termbases? 1. **Consistency**: Same term = same translation every time 2. **Efficiency**: Don’t look up the same term twice 3. **Quality**: Use approved terminology 4. **Client requirements**: Follow style guides ## Termbase Features in Supervertaler ### Automatic Highlighting Terms from active termbases are highlighted in the source text: * Green background by default * Hover to see the translation * Higher priority = darker shade ### Multiple Termbases Maintain separate termbases for: * Different clients * Different domains (legal, medical, IT) * Different projects ### Priority Levels Assign priority (1-10) to terms: * Priority 1 (highest): Must be used * Priority 5: Standard terms * Priority 10: Optional/suggestions ### Forbidden Terms Mark terms as “forbidden” to flag text that should NOT be translated or should be avoided. ## Creating Your First Termbase 1. Go to the **Termbases** tab 2. Click **+ Create Termbase** 3. Enter a name (e.g., “Client ABC Terminology”) 4. Choose source and target languages 5. Click **Create** ## Adding Terms ### Manually 1. Click on your termbase 2. Click **+ Add Term** 3. Enter source term, target term 4. Optionally add domain, notes, priority 5. Click **Save** ### From Selection 1. Select text in the source column 2. Right-click → **Add to Termbase** 3. Enter the target translation 4. Choose which termbase to add to ### Import from File 1. Go to the **Termbases** tab 2. Click **Import** 3. Select a TSV file (tab-separated: source, target, domain, notes) 4. Watch the progress dialog ## Termbase Settings ### Activation Termbases must be activated to show matches: * ✅ **Read**: Terms are highlighted and shown in lookups * ✅ **Write**: New terms can be added during translation ### Highlight Style Choose how terms appear in the grid: * **Background**: Green background shading * **Dotted Underline**: Subtle underline * **Semibold**: Bold text Go to **Settings → View Settings → Termbase Highlight Style**. *** ## See Also * [Creating Termbases](/workbench/termbases/creating/) * [Importing Terms](/workbench/termbases/importing/) * [Term Highlighting](/workbench/termbases/highlighting/) * [Sending Terms to the AI](/workbench/termbases/ai-injection/) * [TermLens (Inline Terminology)](/workbench/termbases/termlens/) * [Term Extraction](/workbench/termbases/extraction/) # Creating Termbases Termbases help you enforce terminology consistently. ## Create a termbase 1. Open the **Termbases** tab 2. Click **Create Termbase** 3. Give it a name and select languages ## Add terms while translating You can build terminology as you work: * Select text in both Source and Target * Use **Add to Termbase** from the context menu The right-click menu offers several routes: **Add to Termbase** (`Ctrl+Alt+T`, opens the entry dialog), and the quick-adds **Quick Add to Project Termbase** (`Alt+Up`) and **Quick Add to Background Termbase** (`Alt+Down`). ## Similar Term Found – merge as a synonym If the term you are adding shares its source with an existing entry (but has a different target), or shares its target (but a different source), Workbench shows a **Similar Term Found** prompt instead of silently creating a near-duplicate. You can: * **Add as Synonym** – fold the new term into the existing entry as a synonym * **Add & Edit…** – do that, then open the entry editor to review it * **Keep Both** – create a separate entry anyway * **Cancel** – abandon the add The prompt only appears when there is an actual overlap, so the quick-adds stay instant otherwise. Exact duplicates (same source *and* target) are skipped as before. This matches the behaviour of the Supervertaler for Trados plugin. ## Tips * Use a separate termbase per client when terminology differs. * Add high-priority terms first (product names, UI strings). # Term Extraction **Term extraction** sends your project’s source text to your configured AI model and returns a proposed **bilingual project glossary** – source terms paired with translations – which you review, edit, and turn into a project termbase in one step. Note **Requires Workbench v1.10.357 or later.** Earlier versions used a mechanical frequency-based extractor (monolingual, no translations); it has been retired. Extraction now uses the same AI provider and model as your translations, configured under **AI → Settings**. ### Opening the extraction dialogue Go to the **Termbases** tab and click **🔍 Extract Terms** in the button bar beneath the termbase list, next to **+ Create New**. Extraction reads the source segments of the open project, so open a project first – clicking with none open just tells you to. ### Choosing the source text The dialogue offers two sources: * **Use project segments** (default) – extracts from all source segments in the loaded project. * **Paste text manually** – enables the text box below, so you can extract from arbitrary text. Useful for a reference document or a client’s style guide. ### Extraction settings | Setting | Default | What it does | | -------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------ | | **Source Language** | the project’s source language | Tells the model what language the text is in. Free text – any language works. | | **Target Language** | the project’s target language | The language the model translates each term into. | | **Domain / Subject** | blank | Optional hint, e.g. `mechanical engineering` or `sewing machines`. Leave blank and the model infers the domain from the text itself. | Click **🤖 Extract Terms with AI**. Small texts return in seconds; a full-length document (e.g. a complete patent application) takes a minute or so. The dialogue stays responsive while it runs. Note **Your source text is sent to the configured AI provider** – the same one that handles your translations, so this adds no exposure beyond translating the project. If a project must not leave your machine, use a local model (Ollama) as your provider. ### Reviewing the results Results fill a table of term pairs: | Column | Meaning | | ---------- | -------------------------------------------------------------------------------------------------------------- | | **Select** | Tick box controlling whether the pair is added. Every row starts ticked. | | **Source** | The term, in its canonical (dictionary) form. **Editable** – click to correct. | | **Target** | The model’s translation. **Editable.** May be empty where the model was unsure – fill it in or untick the row. | | **Note** | Optional context from the model, such as a domain label or a caveat. | Edit cells directly to fix anything before committing – your edits are what gets saved, not the model’s original answer. To tick or untick many rows at once, select them (Shift/Ctrl+click) and **right-click** → *Untick selected* / *Tick selected*. On very large projects only the first portion of the text (roughly 8–9k words) is analysed, and the results label says so explicitly – nothing is truncated silently. ### Creating the project termbase Click **Create Project Termbase**, then give it a name. The default is ` Terminology`. Supervertaler creates a project-scoped **bilingual** termbase containing every ticked pair and makes it the **Project termbase** – it appears in the Termbases tab with the pink **Project** tick and **Read** enabled, so terms with targets immediately produce [TermLens](/workbench/termbases/termlens/) suggestions. Empty-target entries can be completed later – see [Creating termbases](/workbench/termbases/creating/). Note **One project termbase per project.** If the project already has one, you are asked what to do (v1.10.358+): make the new termbase the project termbase (the existing one is kept as a regular termbase), save the new one as a regular termbase alongside, or cancel. Nothing is deleted in any case. ### Tips * The model’s output is a proposal, not a verdict – review it as you would any AI suggestion, especially target translations of ambiguous terms. * A one-word domain hint noticeably improves precision on specialised texts. * If you build prompts with **AutoPrompt**, note that it already performs glossary extraction as part of prompt generation – the two are complementary: this feature produces a *termbase* you can edit, share, and reuse across sessions. *** ### See Also * [Termbase Basics](/workbench/termbases/basics/) * [Creating termbases](/workbench/termbases/creating/) * [Importing termbases](/workbench/termbases/importing/) * [AI injection](/workbench/termbases/ai-injection/) * [TermLens overview](/workbench/termbases/termlens/) # Term Highlighting When a termbase is active, Supervertaler highlights matching terms in the grid. ## Why it helps * Prevents terminology drift * Speeds up review * Makes it obvious when a preferred term exists ## Tips * If the highlight is too strong or too subtle, adjust it in [View Settings](/workbench/settings/view/). * Tag highlighting and termbase highlighting are separate: tags are for placeholders/formatting, termbases are for terminology. * For a word-by-word terminology view with translations underneath each term, see [TermLens](/workbench/termbases/termlens/). # Importing Terms You can import terminology from common formats such as TSV/CSV (and other supported termbase exports). ## Import steps 1. Open the **Termbases** tab 2. Choose **Import** 3. Select your file ## Tips * Clean your source file (consistent columns) before importing. * Import before batch translation so AI can follow your terminology. Note If you don’t see terms highlighting in the grid, ensure the termbase is enabled (Read/active) in the **Termbases** tab. # TermLens TermLens is Supervertaler’s inline terminology display. It shows the source text of the current segment word by word, with termbase translations directly underneath each matched term. ## How it works When you select a segment, TermLens analyses the source text against all active termbases and displays the result in a visual layout: * **Matched words** appear with their termbase translation underneath * **Unmatched words** are shown in light text so you can read the full source sentence in context * **Project termbase** matches appear in pink; **Background termbase** matches appear in blue * **Non-translatable** terms (if configured) are shown in a distinct style This gives you an at-a-glance overview of every term in the segment that has a termbase entry – without having to hover or click anything. ## Where to find it TermLens appears in two places: 1. **Below the grid** – the “TermLens” tab in the bottom panel (toggle with **View → TermLens Under Grid**) 2. **In the Match Panel** – the right-side panel that also shows TM matches Both instances update simultaneously when you navigate to a new segment. ## Inserting terms You can insert a termbase translation from TermLens into your target text in three ways: ### Click to insert Click any translation shown under a source word. The translation is inserted at the cursor position in the target field. ### Keyboard shortcuts (Alt+1 through Alt+9) Each matched term in TermLens is assigned a numbered badge. Press **Alt+1** to insert the first match, **Alt+2** for the second, and so on. **Double-tap** for terms 10 and above: press **Alt+1, Alt+1** quickly to insert term 11, **Alt+2, Alt+2** for term 22, etc. > **Note:** Alt+0 is reserved for the Compare Panel. TermLens numbering starts at 1. ### Right-click menu Right-click a term in TermLens to: * **Insert** the translation * **Edit** the termbase entry * **Delete** the termbase entry ## On-demand views (popup & picker) In addition to the always-visible panels, two on-demand views show the same matches in a more focused layout. Use them when the docked panel is hidden, on small screens, or when you want a keyboard-only insertion flow. | View | Trigger | Best for | | ---------------------------------------------------------- | ---------------- | ------------------------------------------------------------------------------------------- | | [**TermLens popup**](/workbench/termbases/termlens-popup/) | **Ctrl** tap | Floating mirror of the docked panel; cycle chips with arrow keys, insert with Enter or 1–9. | | [**TermPicker**](/workbench/termbases/termpicker/) | **Ctrl+Shift+P** | Modal tabular grid with #/Source/Target/Termbase columns and expandable synonym sub-rows. | Both pull from the same data the docked panel uses, so the chips / rows you see are identical – just laid out differently. ## Font settings You can customise the TermLens font independently from the grid font: 1. Go to **Settings → View Settings → TermLens Font Settings** 2. Choose font family, size (6–16 pt), and bold/normal weight 3. Changes apply immediately to both TermLens instances ## Tips * Press **F5** to force a refresh if matches appear to be missing. * TermLens respects termbase activation – only terms from activated termbases are shown. * If you have many termbases, designate one as the **Project termbase** (shown in pink) to make its terms stand out. ## TermLens for Trados A standalone version of TermLens is also available as a plugin for **Trados Studio 2024+**. It reads the same SQLite termbase format used by Supervertaler and displays terminology matches directly inside the Trados editor. → [TermLens for Trados on GitHub](https://github.com/michaelbeijer/TermLens) *** ## See Also * [TermLens popup](/workbench/termbases/termlens-popup/) – on-demand floating mirror of the panel (Ctrl tap) * [TermPicker](/workbench/termbases/termpicker/) – tabular grid view of the same matches (Ctrl+Shift+P) * [Termbase Basics](/workbench/termbases/basics/) * [Term Highlighting](/workbench/termbases/highlighting/) * [Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) # TermLens popup The **TermLens popup** is a borderless floating version of the docked TermLens panel for the active segment. It mirrors the panel’s chips, colours, and metadata indicators exactly, but appears at the cursor on demand – designed for keyboard-only term selection on small screens, and for translators who want to insert terms without ever reaching for the mouse.  The TermLens popup floating at the cursor over the active segment, with the current match highlighted (blue ring around the chip). The docked TermLens panel on the right shows the same matches – the popup is its on-demand mirror. ### When to use it * **Small screens / laptops** – keeping the docked TermLens panel always-visible can cost too much vertical space, especially for longer source sentences. The popup gives you the same view on demand and disappears when you’re done. * **Pure-keyboard workflows** – Ctrl tap, cycle, Enter, back to typing. No mouse, no menu hunting. ### Opening and closing | Key | Action | | ----------------------- | ------------------------------------------------ | | **Ctrl** (tap) | Toggle the popup (open if closed, close if open) | | **Esc** | Close without inserting | | Click outside the popup | Close without inserting | | Move the mouse > 4 px | Close (the popup is meant to be transient) | A “Ctrl tap” is a press-and-release of the Ctrl key on its own – no other key in between. Same memoQ-style trigger you might already use for the docked panel’s Insert-by-number flow. ### Cycling between matches When the popup opens, the first chip has a thin blue ring around it – that is the **current chip** that Enter will insert. The cycle skips bare source words; only chips you can actually insert get the highlight. | Key | Action | | ------------------------------ | ------------------------------------------------ | | **Right** / **Down** / **Tab** | Move the current-chip highlight to the next chip | | **Left** / **Up** | Move it to the previous chip | Cycling wraps: from the last chip, Right takes you back to the first. ### Inserting | Key / action | Result | | ------------------ | ------------------------------------------------------------------------------------------------- | | **Enter** | Insert the current chip into the target segment, close the popup, return focus to the target cell | | **1–9** | Insert that-numbered chip directly (same numbering as the docked panel’s Alt+N shortcut) | | **Click any chip** | Insert that chip into the target segment, close the popup, return focus to the target cell | All paths share the same insertion logic – there is no difference between picking by keyboard and picking by mouse. ### Editing a match Press **E** while a chip is highlighted to open the term-entry editor for that entry. The popup closes first so the editor opens with clean focus. The editor is the same dialogue the docked panel’s right-click “Edit Termbase Entry…” menu uses, including the multi-termbase “Editing:” dropdown that lets you switch between sibling entries from other active termbases. ### Showing metadata Press **I** while a chip is highlighted to toggle the sticky metadata popup for that entry – the same hover popup the docked panel shows, with the entry’s synonyms, abbreviations, definition, domain, notes, and URL fields. Press **I** again to dismiss it. ### Visuals The popup uses the same chip rendering, colour scheme, and metadata indicators as the docked TermLens panel: | Chip style | Meaning | | ---------------------------------- | ------------------------------------------------ | | **Pink** background | Project termbase term | | **Blue** background | Background termbase term | | **Amber / yellow** background | Non-translatable term | | **Red** background + strikethrough | Forbidden term | | **Purple** background | Match via abbreviation (shows abbreviation pair) | | **ℹ** corner indicator | Entry has metadata (definition / domain / etc.) | | **≡** corner indicator | Entry has synonyms | | **+N** badge on chip | N more cross-termbase entries available | See the [TermLens overview](/workbench/termbases/termlens/) for the full colour key. ### TermLens popup vs TermPicker Both show the same matches for the active segment. Pick whichever fits your style: | | TermLens popup (Ctrl tap) | [TermPicker](/workbench/termbases/termpicker/) (Ctrl+Shift+P) | | -------- | ------------------------------------------------------ | ------------------------------------------------------------- | | Layout | Source segment with chips underneath each matched word | Sortable, scrollable table | | Best for | Skimming matches in segment context | Many matches that benefit from sorting / typing-to-jump | | Keyboard | Arrow / Tab cycles a highlighted chip | 0–9 jumps directly; Up / Down navigate | | Modality | Modeless – Esc / mouse move / click outside to dismiss | Modal – Esc or Cancel to close | *** ### See Also * [TermLens overview](/workbench/termbases/termlens/) * [TermPicker](/workbench/termbases/termpicker/) * [Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) # TermPicker **TermPicker** is a modal dialogue that lists all matched termbase terms and non-translatables for the current segment in a tabular, keyboard-navigable grid. It is useful when [TermLens](/workbench/termbases/termlens/) shows many matches and you want a quick overview without the chip layout – or when you want to compare alternative translations side-by-side before committing.  The TermPicker dialogue floating over the editor for the active segment, listing every termbase + NT match in a sortable grid. Row 3 is expanded (▾) to show a synonym sub-row underneath. The docked TermLens panel on the right shows the same matches as chips. Note **TermLens and TermPicker are sibling surfaces, not parent/child.** Both consume termbase matches for the current segment, but present them in completely different ways: * **TermLens** shows matches *in context* – the source sentence with terms highlighted in place and translation chips anchored to where each term sits. Best for reading and scanning. * **TermPicker** shows the same matches as a flat sortable list with keyboard-driven Enter-to-insert. Best for quick insertion when you already know which term you want. Underneath both: your termbases. ### Opening TermPicker Press **Ctrl+Shift+P** to open TermPicker. It appears as a modal window above the editor. (P = **P**icker – matches the Trados plugin and follows the VS Code-style command-palette convention.) The shortcut is fully remappable via **Settings → Keyboard Shortcuts** under the `term_picker` action. > Looking for the lone-Ctrl-tap behaviour? That opens the [TermLens popup](/workbench/termbases/termlens-popup/) – a more compact in-context view of the same matches. TermPicker described on this page is the table-based alternative for users who prefer a tabular UI. ### Colour-coded rows Each row is colour-coded by its source: | Colour | Meaning | | ------------------- | -------------------------------- | | **Pink** | Project termbase term | | **Blue** | Background termbase term | | **Yellow / amber** | Non-translatable term | | **Grey** (indented) | Synonym sub-row of the row above | This lets you instantly see where each term comes from and how it should be handled. ### Expandable synonyms Terms with multiple translations (either target synonyms recorded on a single entry, or multiple termbase entries hitting the same source word) display a right-arrow indicator (**▸**) next to the row number. To expand and see all alternative sub-rows: * Select the row and press the **Right arrow** key * The sub-rows appear underneath, indented with a `└` and shown in grey, one per available translation * The indicator switches to **▾** to show the row is expanded Press **Left arrow** to collapse the synonyms again. ### Keyboard navigation TermPicker is designed for fast keyboard use: | Key | Action | | ------------- | -------------------------------------------------------------- | | **0–9** | Jump to that-numbered row; auto-inserts when ≤ 9 total matches | | **Enter** | Insert the selected term and close the picker | | **Esc** | Close the picker without inserting | | **Up / Down** | Navigate between rows (wraps around) | | **Right** | Expand synonyms for the selected row | | **Left** | Collapse synonyms (jumps to parent row when on a sub-row) | **Number-key behaviour:** when the segment has 9 or fewer matches, pressing a digit selects *and* inserts the corresponding row in one keystroke. When there are 10 or more matches, the digit only selects the row – press **Enter** to insert. This guards against unintended auto-inserts when a digit was the first character of a two-digit number you were typing. ### Inserting a term You can insert a term in three ways: * **Double-click** any row to insert that term at the cursor position in the target field * **Press Enter** on the selected row to insert and close * **Click “Insert”** in the dialog footer The selected translation lands at the current cursor position in the target segment. ### Persisted layout TermPicker remembers your preferred size and column widths between sessions, so once you resize it to fit your screen the layout sticks. > TermPicker shows the same matches as TermLens, but in a flat sortable list format that scales better when there are many results. *** ### See Also * [TermLens overview](/workbench/termbases/termlens/) * [TermLens popup](/workbench/termbases/termlens-popup/) * [Termbase Basics](/workbench/termbases/basics/) * [Keyboard Shortcuts](/workbench/editor/keyboard-shortcuts/) # Pdf Rescue ## Overview **PDF Rescue** is a specialised AI-powered OCR tool designed to extract clean, editable text from poorly formatted PDFs. Built into Supervertaler, it uses vision-capable LLM OCR to intelligently recognise text, formatting, redactions, stamps, and signatures–producing professional, translator-ready documents. ### 🎯 The Problem It Solves Have you ever received a PDF translation job where: * The text won’t copy-paste cleanly? * Line breaks are all over the place? * Formatting is completely broken? * Traditional OCR produces gibberish? * Redacted sections show as black boxes? * Stamps and signatures clutter the text? **PDF Rescue fixes all of this.** ### Real-World Success Story > *“I had a client reach out for a rush job–a 4-page legal document that had clearly been scanned badly. Traditional OCR couldn’t handle it, and manual retyping would have taken hours.* > > *I used PDF Rescue’s one-click PDF import, processed all 4 pages with AI OCR, and it produced a flawless Word document that I could immediately start working with. What would have been a multi-day nightmare became a straightforward job I could deliver on time.* > > *I was able to tell my client that I could handle the job–and delivered professional quality. PDF Rescue literally saved a client relationship.”* > > – Michael Beijer, Professional Translator *** ## ✨ Key Features ### 1. 📄 **One-Click PDF Import** * **No external tools needed** - Import PDFs directly * **Automatic page extraction** - Each page saved as high-quality PNG (2x resolution) * **Persistent storage** - Images saved next to source PDF in `{filename}_images/` folder * **Client-ready** - Images can be delivered to end clients if needed ### 2. 🧠 **Smart AI-Powered OCR** * **Vision-capable LLM OCR** - High accuracy OCR * **Context-aware** - Understands document structure and formatting * **Intelligent cleanup** - Fixes line breaks, spacing, and formatting issues * **Redaction handling** - Inserts descriptive placeholders like `[naam]`, `[bedrag]` in document language * **Stamps & signatures** - Detects and describes non-text elements: `[stempel]`, `[handtekening]` ### 3. 🎨 **Optional Formatting Preservation** * **Markdown-based** - Uses `**bold**`, `*italic*`, `__underline__` * **Toggle on/off** - User-controlled via checkbox * **Clean output** - Markdown converted to proper formatting in DOCX export * **Visual preview** - See formatting markers before export ### 4. 📊 **Batch Processing** * **Process selected** - Work on individual images * **Process all** - Batch process entire document * **Progress tracking** - Visual progress bar and status updates * **Skip processed** - Already-processed images are skipped (unless re-selected) ### 5. 📝 **Comprehensive Logging** * **Activity log integration** - All operations logged with timestamps * **PDF import progress** - Each page extraction logged * **OCR processing** - Per-image processing logged * **DOCX export** - Export operations tracked ### 6. 👁️ **Full Transparency** * **“Show Prompt” button** - View exact instructions sent to AI * **Configuration display** - See model, formatting settings, max tokens * **No black boxes** - Complete visibility into AI processing ### 7. 📊 **Professional Session Reports** * **Markdown format** - Clean, readable documentation * **Complete configuration** - All settings recorded * **Processing summary** - Table of all images and status * **Full extracted text** - All OCR results included * **Statistics** - Character/word counts and averages * **Supervertaler branding** - Professional client-ready reports ### 8. 💾 **Flexible Export Options** * **DOCX export** - Formatted Word documents with optional bold/italic/underline * **Copy to clipboard** - Quick text extraction * **Session reports** - Professional MD documentation ### 9. 🚀 **Standalone Mode** Can run independently outside Supervertaler: ```bash python modules/pdf_rescue.py ``` Full-featured standalone application with all capabilities. *** ## 🎯 Workflow ### Quick Start (5 Steps) 1. **Open PDF Rescue** - Open the **Tools menu** at the top of the window → **🔍 PDF Rescue**. The tool opens in its own window. 2. **Import PDF** - Click ”📄 PDF” button, select your badly-formatted PDF 3. **Check formatting option** - Leave “Preserve formatting” checked (default) 4. **Process** - Click ”⚡ Process ALL” to OCR all pages 5. **Export** - Click ”💾 Save DOCX” to create Word document **That’s it!** You now have a clean, editable Word document ready for translation. *** ### Detailed Workflow #### Step 1: Import Your PDF **Method 1: Direct PDF Import** (Recommended) ```plaintext Click: 📄 PDF button → Select PDF file → Automatic page extraction to {filename}_images/ folder → All pages added to processing queue ``` **Method 2: Manual Image Import** ```plaintext Click: 📁 Add Files → Select individual images OR Click: 📂 Folder → Select folder with images ``` **Result**: All images listed in left panel with ✓ status indicators *** #### Step 2: Configure Settings **Model Selection** (vision-capable models, grouped by provider): * **OpenAI**: `gpt-5.5` (Recommended - flagship), `gpt-5.4-mini` (budget option) * **Claude**: `claude-sonnet-4-6`, `claude-haiku-4-5-20251001`, `claude-opus-4-8` * **Gemini**: `gemini-3.1-flash-lite`, `gemini-2.5-pro`, `gemini-3.1-pro-preview` **Formatting Option**: * ✓ **Preserve formatting (bold/italic/underline)** - Enabled by default * Unchecked = Plain text output only **Extraction Instructions**: * Default instructions optimized for badly formatted PDFs * Handles redactions, stamps, signatures automatically * Can customize if needed (advanced) * Click **“👁️ Show Prompt”** to see exact AI instructions *** #### Step 3: Process Images **Option A: Process Selected** ```plaintext 1. Select image(s) in list 2. Click: 🔍 Process Selected 3. View result in preview pane ``` **Option B: Process All** (Recommended) ```plaintext 1. Click: ⚡ Process ALL 2. Confirm batch processing dialog 3. Watch progress bar 4. All pages processed automatically ``` **Processing Details**: * Each image sent to the OCR model * Text extracted with context awareness * Formatting detected (if enabled) * Redactions/stamps/signatures handled * Results stored in memory * ✓ indicator appears when processed *** #### Step 4: Review & Export **Review Extracted Text**: * Click any processed image in list * Preview pane shows extracted text * Formatting shown as markdown (`**bold**`, `*italic*`, etc.) * Verify quality before export **Export Options**: 1. **💾 Save DOCX** (Primary export) * Formatted Word document * Markdown converted to proper formatting * One page per document page * Page headers with filenames * Ready for translation work 2. **📋 Copy All** * All text to clipboard * Includes page separators * Quick paste into any application 3. **📊 Session Report** * Professional markdown documentation * Complete configuration record * All extracted text included * Statistics and metadata * Client-ready deliverable # Statistics (Analyse Against TM) The **Statistics** tool analyses the document you have open against one or more of your translation memories and produces a match breakdown – the same kind of report you get from the “Analyze Files” step in Trados Studio or memoQ. It tells you, before you start, how much of the job is already covered by your TMs and how much is genuinely new, so you can scope the work and quote accurately. ## Where to find it * Open the **Tools menu** → **📊 Statistics (Analyse Against TM)…** You need a project open with segments. Press **F1** (or the **?** in the top-right of the dialog) to return to this page at any time. ## ⚡ Quick Count (no project needed) When you just want a fast count and don’t want to set up a project, use **Tools → ⚡ Quick Count…** instead: 1. Browse to **one or more files**. Supported: **DOCX** (plus IDML, HTML, XLIFF, PO, XLSX, PPTX via Okapi) and the CAT bilingual formats **Trados `.sdlxliff`** and **memoQ `.mqxliff`**. 2. The same Statistics dialog opens – pick your TMs and matching depth, then **Analyse**. DOCX files are sentence-segmented through Okapi exactly like a normal import, so the numbers match the project-based tool. If a file can’t be read it’s reported on its own and the rest are still counted. The language pair used for segmentation/matching is the open project’s, or your last-used import pair if no project is open. Note **Selecting several files:** in the file browser, Ctrl-click (or Shift-click) to select multiple files in the *same folder*, then click **Open**. The native dialog can’t select across different folders at once. ### Per-file breakdown When you analyse **more than one file** (Quick Count with several files, or a multi-file project), each translation memory’s result shows a combined **All files** total followed by a **per-file** table, so you can see how each file contributes. The breakdown also appears in the HTML/Excel/CSV exports. Single-file analyses just show the one total. ## How to use it 1. Tick one or more translation memories to analyse against. The TMs already activated for the current project are ticked for you. * **Leave every TM unticked** to get a plain word count plus internal repetitions only (no TM lookup). 2. Choose a **Matching depth** (see below). 3. Click **Analyse**. The analysis runs in the background – you can cancel it at any time. Results appear per TM as each one finishes. 4. Optionally click **Export…** to save the report as **HTML**, **Excel (.xlsx)**, or **CSV**. ## Matching depth The fuzzy-match pass is the slow part on a large TM, so you can trade thoroughness for speed: | Depth | What it does | | ---------------------- | --------------------------------------------------------------------------------------------------------- | | **Standard** | Exact matches plus fuzzy matches down to 75%. The default – fast and covers almost all reusable material. | | **Thorough** | Exact plus fuzzy down to 50%. Fills the lower fuzzy bands; a little slower. | | **Exact matches only** | Skips the fuzzy pass entirely. Near-instant, even on a TM with hundreds of thousands of entries. | Behind the scenes the fuzzy search uses the TM’s full-text index to look only at the most relevant candidates (no sub-segment/fragment search), which is conceptually the same as Trados Studio’s “Optimized Performance” option. ## What the match types mean | Type | Meaning | | ------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------- | | **Repetitions** | Source text that repeats earlier in the document. The first occurrence is counted under a match band; the repeats land here (translate once, reuse). | | **101% (Context Match)** | An exact match whose surrounding context also matches the TM – the safest reuse, normally needs no editing. | | **100%** | An exact match of the source text in the TM (context not checked). | | **95%–99%** | Very high fuzzy match – usually a tiny edit. | | **85%–94%** | High fuzzy match – minor editing expected. | | **75%–84%** | Medium fuzzy match – noticeable editing expected. | | **50%–74%** | Low fuzzy match – often faster to retranslate than to fix. | | **No match** | No usable TM match – translate from scratch. | Each row reports the number of **segments**, **words**, **characters** (tags excluded), **tags**, and the **percentage** of total words. The exported report includes this legend and the project name. ## Related * [Translation memory](/workbench/translation-memory/basics/) * [Fuzzy matching](/workbench/translation-memory/fuzzy-matching/) * [Managing TMs](/workbench/translation-memory/managing-tms/) # Superbrowser (Multi-Chat AI Browser) Superbrowser puts ChatGPT, Claude, and Gemini side by side in one window – three resizable browser columns, each with its own persistent login – so you can put the same question to all three and compare the answers without switching tabs or juggling browser windows. Open it from **Tools → 🌐 Superbrowser…**. It opens in its own window, so it can live on a second monitor next to your translation work.  One question, three opinions – ChatGPT, Claude, and Gemini answering “What is the best CAT tool in 2026?” side by side. ### How it works * **Three columns, three services** – ChatGPT (left), Claude (centre), Gemini (right), each in a full embedded browser. Drag the dividers to resize. * **Persistent logins** – each column keeps its own isolated browser profile, so you sign in once per service and stay signed in between Supervertaler sessions. Profiles are stored under `workbench/superbrowser_profiles/` in your Supervertaler data folder. * **Custom URLs** – click **Show Configuration** (top right) to point any column at a different address – a specific chat session, a project conversation, or another service entirely – and apply all three at once with **Update URLs**. ### Why compare models? Different models genuinely differ on translation questions – terminology choices, register, how they handle an ambiguous source. Asking all three at once tells you in seconds whether an answer is a consensus or one model’s opinion, which is exactly when a second opinion is worth having. Note Superbrowser was removed in v1.9.385 during a round of simplification, and restored in v1.10.365 after a user asked for it back. If you used it before, your logins should still be there – the removal deliberately left the profile folder on disk, and the restored tool picks it straight back up. Caution The columns are real browser sessions, so each service’s own sign-in flow applies. If a provider objects to an embedded browser during login (Google is occasionally strict about this), complete the sign-in once in the column and it will be remembered from then on. # TMX Editor Supervertaler includes a built-in TMX editor for inspecting and editing TMX translation memories. ## Where to find it * Open the **Tools menu** at the top of the window → **✏️ TMX Editor**. The editor opens in its own window. ## What you can do * Open, edit, and save TMX files. * Search and filter by source/target text. * Edit TMX header metadata. * Run basic validation and view statistics. * Perform bulk operations (for example: delete entries, copy source → target). ## Common workflows ### Clean up a TMX before importing 1. Open the TMX in **✏️ TMX Editor**. 2. Fix any obvious formatting issues (wrong language, empty segments, etc.). 3. Save the TMX. 4. Import it into your project via [Importing TMX files](/workbench/translation-memory/importing-tmx/). ### Remove unwanted tags If you’re trying to simplify a TMX that contains formatting or CAT-tool tags, you can remove them before importing. Note TMX is just XML – some tags are real inline markup (TMX/XLIFF-style), others are literal text like `<b>...</b>`. Cleaning tags can improve matching, but it can also remove important formatting. If you’re unsure, test on a copy first. ## Related * [Importing TMX files](/workbench/translation-memory/importing-tmx/) * [Translation memory](/workbench/translation-memory/basics/) # Translation Memory Basics Translation Memory (TM) helps you reuse previous translations. ## What is Translation Memory? A Translation Memory stores pairs of source and target text: | Source | Target | Match % | | ------------------- | --------------------- | ------- | | “Save the file" | "Sla het bestand op” | 100% | | “Save the document" | "Sla het document op” | 85% | When you translate new text, the TM finds similar segments. ## How TM Works 1. **You translate** a segment 2. **TM stores** the source + target pair 3. **Later**, when similar text appears: * TM finds matches * Shows them in the Translation Results panel * You can insert or adapt them ## Match Types | Type | Match % | Description | | ---------------- | ------- | ------------------------------------------- | | **Exact** | 100% | Identical source text | | **High Fuzzy** | 90-99% | Minor differences (numbers, capitalization) | | **Medium Fuzzy** | 75-89% | Some words different | | **Low Fuzzy** | 50-74% | Significant differences | ## Benefits ### Save Time Don’t translate the same sentence twice. TM automatically suggests previous translations. ### Consistency Same source = same translation. Important for technical documentation, UI strings, and legal text. ### Cost Savings Clients often pay less for TM matches: * 100% match: Lowest rate * Fuzzy match: Reduced rate * New text: Full rate ## TM in Supervertaler ### Translation Results Panel When you select a segment: 1. TM searches for matches 2. Results appear in the panel on the right 3. Shows match percentage, source, and target 4. Double-click to insert ### Using Matches | Action | How | | --------------- | ------------------------------------------------- | | Insert match | Double-click, or select it and press `Ctrl+Space` | | Copy match | Right-click → Copy | | View in context | Right-click → View in TM | ### Building TM Your TM grows as you translate: 1. Translate a segment 2. Confirm with `Ctrl+Enter` 3. The pair is saved to active TM ## Multiple TMs You can have multiple TMs: * **Project TM**: For the current project * **Client TMs**: One per client * **Master TM**: All your translations ### TM Priority When multiple TMs match: * Higher priority TMs shown first * Can reorder in the **TMs** tab ### Sending segments to several TMs at once To push your confirmed translations into more than one TM in a single run, use **Bulk Operations ▸ Update active TMs**. Under **Target Translation Memories**, tick every TM you want to update – the segments are written to all of them in one pass, with a per-TM summary of how many were sent to each. *** ## See Also * [Creating & Managing TMs](/workbench/translation-memory/managing-tms/) * [Importing TMX Files](/workbench/translation-memory/importing-tmx/) * [Fuzzy Matching](/workbench/translation-memory/fuzzy-matching/) # Fuzzy Matching Fuzzy matching finds similar segments (not just exact duplicates). ## How to use it * As you navigate, Supervertaler searches your TMs for similar source text. * Matches are scored by similarity. ## When to trust a match * **High scores** are often safe to insert as a starting point. * **Mid/low scores** can still be useful, but should be treated as suggestions. ## Tips * Always review fuzzy matches before inserting. * For formatted text, preserve tags when inserting matches. # Importing TMX Files TMX is the common exchange format for translation memories. ## Import steps 1. Open the **TMs** tab 2. Choose **Import TMX** 3. Select your `.tmx` file Note Import your TM(s) before batch translation to maximize reuse. ## Tips * Import TMs before batch translation to maximize reuse. * If matches don’t show up, verify the TM language pair. # Creating & Managing TMs Translation Memories (TMs) help you reuse past translations. ## Add or create a TM 1. Open the **TMs** tab 2. Add an existing TM (or create a new one) 3. Enable **Read** to use it for matches 4. Enable **Write** if you want new translations saved into it ## Using TM matches * TM matches appear automatically as you navigate. * You can insert matches quickly: * **Ctrl+Space** inserts the currently selected match * Double-click a match in the panel to insert it ## Tips * Keep separate TMs per client or domain if needed. * Ensure the TM language pair matches your project. * If a segment contains tags/placeholders, keep them intact when inserting a match. # Trados Sdltm You can attach a Trados Studio Translation Memory (`.sdltm` file) directly to Supervertaler and consult it for matches – without exporting it to TMX first, and without closing Trados. ## When to use this The typical scenario: you’re translating a project in Trados Studio with a working TM. You’d like to consult that same TM from inside Supervertaler – maybe to use its concordance, run AI translations against it, or work on a different project that shares terminology with the Trados project. Rather than exporting to TMX every few minutes, you point Supervertaler at the `.sdltm` directly and let it stay in sync. ## Attach a Trados TM 1. Open **TMs → TM List**. 2. Click **🔗 Attach Trados TM** (next to **📥 Import TMX**). 3. Pick the `.sdltm` file. 4. Confirm the dialog – it shows the TM’s languages, name, and translation-unit count. 5. Optionally edit the display name (defaults to the full filename including `.sdltm` so it’s easy to spot in the TM list). 6. Watch the progress dialog as Supervertaler mirrors the TUs (≈ 5 seconds for a 13 K-TU TM). The TM is created **read-only** by default. Supervertaler never writes back to your `.sdltm`; that file remains the source of truth and stays under Trados’s exclusive control for writes. ## How sync works Once attached, the mirror **stays in sync with the live `.sdltm`**. Every 5 seconds Supervertaler checks the file’s modification time. When Trados saves a new or modified TU and the file timestamp changes, Supervertaler pulls in just the delta – not a full re-read – and updates the TM’s entry count automatically. The mtime check itself is essentially free, so idle ticks cost nothing. You can keep working in Trados; new TUs you confirm there appear in Supervertaler within a few seconds. ## Tag preservation Trados-style inline tags are preserved as Supervertaler’s `...` markers on the way in: | Trados | Supervertaler | | --------------------- | ------------- | | Start tag, ID 116 | `<116>` | | End tag, ID 116 | `` | | Standalone tag, ID 12 | `<12/>` | This is the same format Supervertaler uses when it imports SDLXLIFF segments directly, so a TM hit drops cleanly into the editor grid alongside your working segments. The tag IDs in the TM hit won’t necessarily match the IDs in the segment you’re translating – TM `<116>` might be your current `<11>`. Structure (start/end pairs, count) is preserved, but you’ll still need to renumber tags on insertion if the IDs differ. ## Refreshing manually If you’d rather refresh on demand than rely on the 5-second timer, click **🔗 Attach Trados TM** again on the same `.sdltm`. You’ll be asked “TM already attached – replace?”; click **Yes** and Supervertaler wipes the existing entries and re-mirrors from disk in one go. ## Limitations * **Read-only**: Supervertaler never writes back to the `.sdltm`. New translations you confirm in the Workbench go to your normal Supervertaler TMs, not back into the Trados file. * **Tag IDs differ**: as noted above, the numeric IDs on tags in TM hits may differ from the IDs in your current segment. Tag *structure* is preserved. * **Match scores differ slightly**: Supervertaler uses Python’s SequenceMatcher; Trados uses its own token-aligned algorithm. Ranking of matches is similar; exact percentages can be a few points apart. * **Concurrent use is safe**: opening the `.sdltm` while Trados has it open is fine – Supervertaler reads via SQLite’s URI read-only mode, and Trados uses WAL. # API Connection Problems If AI translation isn’t working, this page helps you diagnose provider/API issues. ## Common causes * Missing or invalid API key * No internet connection * Provider rate limits or quota limits * Wrong model selected ## Quick checks 1. Verify your API key in [Setting Up API Keys](/workbench/get-started/api-keys/) 2. Confirm you selected a **provider** and **model** in Settings (LLM/AI settings) 3. Try translating a single short segment (`Ctrl+T`) 4. Check whether your account has credits/quota ## Common errors and fixes | Symptom / message | Likely cause | What to do | | ----------------------------------------- | ----------------------------------------------------- | -------------------------------------------------------------------------------------------- | | “Invalid API key” / authentication failed | Key is wrong or has extra whitespace | Re-paste the key; make sure there are no leading/trailing spaces; Save and restart if needed | | “Rate limit exceeded” | Provider is throttling requests | Wait 1–2 minutes; reduce batch size; try a different model | | “Quota/credits exceeded” | Account has no credits or billing disabled | Check provider dashboard; add credits/enable billing | | “Model not found” | Selected model name not available | Pick a supported model in Settings; update if the provider changed model names | | “No response” / empty translation | Transient provider failure, network issue, or timeout | Try single-segment translation; retry the segment; try a different model/provider | | Connection errors / timeouts | Network/VPN/firewall/proxy issues | Try another network; disable VPN; allow Python/Supervertaler through firewall | ## Tips for reliable translation * Start with **single-segment** translation to verify your setup before running batch. * If batch translation returns empty segments, enable the retry option in the batch dialog. * If your segments contain tags/placeholders, add a prompt rule to preserve them exactly. ## Error messages If you see an error dialog, copy the message and include it when asking for help. Include: * Provider + model * Whether single-segment translation works * The exact error text # Common Issues Solutions to frequently encountered problems. ## Startup Issues ### Application won’t start **Symptoms:** Double-click does nothing, or window briefly appears then closes. **Solutions:** 1. **Run from command line** to see error messages: ```bash python Supervertaler.py ``` 2. **Reset UI preferences** (corrupted window state): * Delete `user_data/ui_preferences.json` * Restart the application 3. **Check dependencies**: ```bash pip install -r requirements.txt ``` 4. **Verify Python version** (needs 3.10+): ```bash python --version ``` ### “Module not found” error Install the missing module: ```bash pip install ``` Or reinstall all dependencies: ```bash pip install -r requirements.txt --force-reinstall ``` *** ## Import Problems ### ”Cannot read file” error * Close the file in other programs (Word, Excel, etc.) * Check if the file is read-only * Try copying the file to a different location ### memoQ bilingual shows no segments * Ensure you exported as **Bilingual DOCX** (table format) * Check the file in Word to verify it has a source/target table ### Trados package fails to extract * The SDLPPX might be corrupted * Re-export from Trados Studio * Check if the package includes all required files ### Encoding errors (garbled text) * Open the source file in Notepad++ to check the detected encoding; if it shows ANSI/Windows-1252 with mojibake, use *Encoding → Convert to UTF-8* and re-save. * For more stubborn cases, run [`ftfy`](https://pypi.org/project/ftfy/) on the file from the command line. * Re-export from the source tool with UTF-8 encoding when possible. *** ## Translation Issues ### AI translation returns empty 1. Check your API key is valid 2. Verify you have credits with the provider 3. Check internet connection 4. Try a different model ### ”Rate limit exceeded” error * Wait 1-2 minutes and try again * Reduce batch size * Upgrade your API plan ### Wrong translation language * Check your prompt specifies the correct language pair * Verify source/target languages are set correctly in project settings ### Tags are removed or moved Add explicit instructions to your prompt: ```plaintext Keep all formatting tags like {1}, , in exactly the same positions in the translation. ``` *** ## Reimporting Issues ### Segments don’t match on reimport **Cause:** Segment structure changed. **Solution:** * Don’t merge or split segments in Supervertaler. * Export the matching format for your CAT tool/workflow. * If you’re working from a bilingual table, don’t modify the table structure in Word. ### Formatting lost on reimport **Cause:** Tags/placeholders weren’t preserved. **Solution:** * Verify tags in Tag view before exporting. * Ensure tags are balanced and not renumbered. * Run CAT tool QA after import to catch tag issues early. ### TM matches not appearing **Cause:** TM not loaded, disabled, or language mismatch. **Solution:** * Open the **TMs** tab. * Ensure the TM is added and **Read** is enabled. * Verify source/target language pair matches your project. *** ## Export Problems ### Exported DOCX has no translations * Make sure you translated the segments (target column isn’t empty) * Check you’re exporting the correct format * Verify the segments are confirmed ### ”Source file not found” on export The original imported file was moved or deleted. * Use **Project → Export → 🔗 Relocate Source Folder** to point to the new location * Or re-import the source file ### Formatting lost after round-trip * Keep all inline tags in your translations * Don’t modify the structure of bilingual tables * Check CAT tool import settings *** ## Performance Issues ### Application is slow 1. **Pick a smaller Per page size** above the grid (the grid shows all segments by default; try 100 or 50) 2. **Disable spellcheck** if not needed (Settings → View) 3. **Close other heavy applications** ### Large files take forever to import * Very large files (10,000+ segments) may take time * Consider splitting into smaller files * Use multi-file import for better organization *** ## Spellcheck Issues ### Spellcheck not working 1. Check spellcheck is enabled: **Settings → View Settings → Spellcheck** 2. Verify the correct language is selected 3. For Hunspell, ensure dictionaries are installed ### Wrong language being checked Go to **Settings → View Settings → Spellcheck** and select the correct target language. ### Red underlines appear everywhere The spellcheck language might not match your target language. Or you might need to add technical terms to your dictionary. *** ## UI Issues ### Dark mode colors look wrong Some widgets apply styles when becoming visible. Try: * Switch themes back and forth * Restart the application ### Window opens off-screen Delete `user_data/ui_preferences.json` to reset window position. ### Fonts look different/wrong Go to **Settings → View Settings** and select your preferred font family. *** ## Still Having Issues? 1. Check the [GitHub Issues](https://github.com/Supervertaler/Supervertaler-Workbench/issues) for known bugs 2. Open a new issue with: * Your OS and Python version * Steps to reproduce the problem * Error messages (if any) * Screenshots (if helpful) # Import/Export Errors This page covers common problems when importing or exporting files. ## “Cannot read file” / import fails **Common causes:** * The file is open in Word (or another app) * The file is read-only or in a protected location * The file format doesn’t match the workflow **Fix:** 1. Close the file everywhere (Word, CAT tool editors, preview panes) 2. Copy it to a simple path (for example `C:\Temp\`) and try again 3. Re-export from the CAT tool using the recommended bilingual/package format ## memoQ bilingual shows no segments **Cause:** Export format doesn’t contain the expected bilingual table. **Fix:** * Re-export from memoQ as **Bilingual DOCX** in a **two-column table** format. * Open the DOCX in Word and confirm it really contains a Source/Target table. ## Trados SDLPPX fails to extract **Common causes:** * Corrupt/partial package * Unsupported package structure **Fix:** * Ask for a fresh export from Trados Studio. * Ensure the package includes all required files. ## Garbled characters / encoding issues **Cause:** Text encoding problems coming from the source file or export. **Fix:** * Re-export from the source tool with a modern Unicode/UTF-8-friendly path when possible. * For Latin-1/Windows-1252 mojibake (e.g. “é” appearing where “é” should be), open the file in a text editor that supports re-interpreting the encoding (Notepad++ has *Encoding → Convert to UTF-8*) or run `ftfy` on it from the command line. ## Segments don’t match on reimport **Cause:** Segment structure changed. **Fix:** * Don’t merge or split segments in Supervertaler * Export using the matching CAT format * Avoid deleting placeholder/tag-only segments ## “Source file not found” during export **Cause:** The original source file/folder was moved after import. **Fix:** * Use **Project → Export → 🔗 Relocate Source Folder** and point it to the new location. * If the original source is gone, re-import the project from the correct source. ## Exported file has no translations **Cause:** Targets are empty or the wrong export format was chosen. **Fix:** * Verify the Target column contains translations. * Export the matching format for the workflow you imported. ## Formatting lost on reimport **Cause:** Tags not preserved. **Fix:** * Verify tags are balanced (for example `text`) * Don’t delete CAT placeholder tags * Re-export using the correct CAT workflow ## Bilingual table reimport fails **Cause:** Bilingual tables are great for review, but not always suitable for CAT tool reimport. **Fix:** * Prefer the dedicated CAT exchange formats (memoQ/Trados/Phrase/CafeTran). * If you must use a bilingual table, don’t edit the table structure in Word. # Linux-Specific Issues Supervertaler is Linux compatible, but Windows is the primary development platform. ## Spellcheck dictionaries If spellcheck isn’t working, you may need to install Hunspell dictionaries for your language: ```bash sudo apt install hunspell-en-us ``` If the dictionary package for your language exists, install it using the language code you need (for example `hunspell-de-de`, `hunspell-nl`, etc.). ## Stability tips If you encounter instability: * Disable spellcheck temporarily * Disable semantic memory features temporarily * Use a smaller project to isolate the issue ## Crashes / memory access violations Some native dependencies (spellcheck backends, tokenization libraries) can crash the Python process on certain Linux setups. If you see random crashes (segfaults) when interacting with the grid: 1. Disable spellcheck and restart 2. Retry on a smaller project If that stabilizes the app, re-enable features one by one. # Performance Tips Supervertaler is designed to stay responsive on large projects, but performance can vary with project size and enabled features. ## Tips * Pick a smaller **Per page** size for very large projects (the grid shows all segments by default; projects over 2000 segments auto-paginate at 500). See [Pagination](/workbench/editor/pagination/) * Keep only the needed TMs and termbases enabled * If semantic search is enabled, allow indexing to complete ## Quick wins when it feels slow 1. Disable spellcheck temporarily (Settings → View Settings) 2. Close other heavy apps (browsers with many tabs, IDE builds, etc.) 3. Restart Supervertaler and reopen the project ## Large projects Very large files (thousands of segments) can stress any UI grid. * Prefer pagination. * Consider splitting source documents or using multi-file projects. ## If it feels slow Try restarting the app and reopening the project. If performance is still poor, note: * Project size (segment count) * Whether spellcheck is enabled * Which CAT format you imported # Voice **Voice** is Supervertaler’s voice command and dictation engine. It lets you control any application on your computer – Trados, memoQ, Word, or anything else in the foreground – using your voice, while Supervertaler Workbench stays running in the background. Open it via the **🎤 Voice** top tab in Workbench, the tray icon’s **Open Voice** entry, or press **Ctrl+Alt+O** to toggle Always-On listening from anywhere on your computer.  *** ## Three modes ### Always-On listening Always-On runs a continuous microphone stream in the background. When you speak, Voice detects speech via amplitude-based VAD (voice activity detection), captures the utterance, and hands it to the active recognition engine. **With the Vosk engine** *(default)* the recogniser only emits text for phrases in your command list – anything else is silently dropped as `[unk]`. So Vosk Always-On is “commands only” by design: you can leave it on all day, talk to colleagues, take phone calls, etc., and only matching command phrases will trigger actions. **With faster-whisper or OpenAI Whisper API** every utterance is transcribed in full. If it matches a command the action fires; if not (and “Listen for commands only” is off), the transcribed text is typed into whichever window is in the foreground. **To start:** click **▶ Start Always-On** in the Voice tab, or press **Ctrl+Alt+O** from any application. A red mic icon appears in the system tray while Always-On is active. **To stop:** click **⏹ Stop Always-On** or press **Ctrl+Alt+O** again. **Focus matters:** Voice sends keystrokes and text to whichever window is currently focused. After starting Always-On, click into Trados, Word, or your browser before you speak. ### Push-to-Talk dictation (Ctrl+Shift+Space) Press **Ctrl+Shift+Space** (the default dictation hotkey – ⌘⇧Space on macOS; works globally, from any application, and is configurable in **Settings → Keyboard Shortcuts**) to record a single utterance for free-form running-text dictation. A small ”🎤 Listening…” toast appears in the top-right of the screen so you know the recording is live; it goes away again when you stop. Recording stops when you release the key (in hold-to-talk mode) or when you press the trigger again (in toggle mode). The transcribed text is then typed at the cursor position. **Always-On + push-to-talk coexist.** If Always-On is running when you trigger push-to-talk, Voice pauses the always-on listener for the duration of the recording, runs the dictation, then resumes Always-On automatically. So you get free continuous Vosk command recognition all day *plus* a hotkey for occasional running-text dictation, without having to manually toggle Always-On off and on. **Push-to-talk modes** (configurable in the Push-to-Talk settings): * **Toggle** (default) – press once to start, press again to stop * **Hold-to-talk** – hold the key to record, release to stop. *Note: hold-to-talk only works if you rebind dictation to a non-global (in-app) key. The default global hotkey always uses Toggle mode (Windows can’t reliably deliver key-up events across processes for global hotkeys).* ### Push-to-Talk for commands (Ctrl+Alt+V) – v1.10.193 Press and **hold** **Ctrl+Alt+V** (the default; configurable in **Settings → Keyboard Shortcuts**) to temporarily activate the command listener for the duration of the hold. Release the key to stop it again. Works globally, from any application. This is a third mode that sits between the two above: * Always-On is the **toggle** version of command listening – mic open continuously, listens for commands all day. * Command Push-to-Talk is the **hold** version – mic open only while you hold the chord, so the rest of the time the microphone is genuinely free for other applications. **When to use this mode:** * You also use an external dictation app (Wispr Flow, Dragon, macOS Dictation, etc.) for running-text dictation and don’t want Supervertaler’s always-on mic competing for the audio stream. * You only need voice commands occasionally – pressing a hotkey when you want to issue one is less intrusive than leaving the mic open all day. * You’re on a laptop battery-conscious about the always-on Vosk model running 24/7. **Coexistence with the toggle mode:** if Always-On is already running when you press Ctrl+Alt+V, the hotkey is a no-op – it won’t restart what’s already going, and releasing it won’t stop Always-On either (we never touch what we didn’t start). So the two modes don’t fight each other; you can use whichever feels right for the moment. **Platform notes:** release detection on Windows uses `GetAsyncKeyState` polling (same mechanism as the dictate PTT). On macOS / Linux, the listener stays running until you press Ctrl+Alt+O or click ⏹ Stop Always-On – it doesn’t auto-stop on key release. Lift to a manual toggle there. ### Pause Always-On for external dictation – v1.10.246 The opposite trade-off to Command Push-to-Talk: keep Always-On running **permanently**, but have it step off the microphone for the moments you’re dictating into an **external** tool (Wispr Flow, Dragon, macOS Dictation, …). You bind one of *your* keys – the same key you press to start your external dictation – and Always-On pauses while it’s engaged, then resumes. So your voice commands stay available all day, and the two never fight over the mic. The key is **recorded, not typed**, so it works with keys you can’t express as text – including media keys like **fast-forward**, which many people use to trigger their dictation tool. **To set it up** – Voice tab → **⏸️ Pause Always-On for external dictation**: 1. Click **Record key**, then press the key you use for your external dictation tool. The label shows what was captured (e.g. *Media Next / Fast-Forward*). 2. Choose a **mode**: * **Hold** *(default)* – Always-On pauses only while you hold the key and resumes the instant you release it. Pair this with **hold-to-talk** tools like Wispr Flow: hold your key → speak → release, and Always-On is live again. * **Toggle** – press once to pause, press again to resume. Use this if your tool starts/stops dictation on a single tap. It works **globally** (the Workbench doesn’t need to be focused) and the key is observed *passively* – your external tool still receives it normally. Press detection and release both come from the same low-level hook used by Command Push-to-Talk. Note **Which to use – this or Command Push-to-Talk (Ctrl+Alt+V)?** They solve the same problem from opposite ends. Command Push-to-Talk keeps Always-On **off** and listens for commands only while you hold its chord. The pause hotkey keeps Always-On **on** and only pauses it while you hold *your* key. Pick the pause hotkey if you want commands available most of the time and just need to duck out of the mic during external dictation. *** ## Voice commands Voice commands execute specific actions when you speak a trigger phrase. They can type text, press keyboard shortcuts, run AutoHotkey scripts, or call internal Workbench functions. ### The commands table The commands table (right side of the Voice tab) lists all your configured commands. | Column | Description | | -------- | -------------------------------------------------------------------------- | | ☑ | Enable/disable checkbox – uncheck to silence a command without deleting it | | Phrase | The primary trigger word or phrase | | Aliases | Alternative phrases that also trigger the command | | Type | Command / Keystroke / AHK Script / AHK Inline | | Action | What happens when the phrase is recognised | | Category | Organisational label (Navigation, Editing, etc.) | ### Enabling and disabling commands * **Single command** – click the checkbox in the first column * **All commands** – click the checkbox column header to toggle all at once (enables any disabled, or disables all if all are already active) * **Multiple commands** – select rows with **Shift+Click** or **Ctrl+Click**, then right-click and choose **✅ Activate** or **⬜ Deactivate** Disabled commands are greyed out and are skipped during recognition. Their settings are preserved. ### Editing a command Double-click any row to open the Edit Voice Command dialog. You can change the phrase, aliases, action type, and action value. You can also use the **Edit** button in the toolbar above the table. ### Adding a command Click **+ Add** above the table. Choose a command type: * **Command** – calls an internal Workbench action (confirm segment, next segment, etc.) * **Keystroke** – sends a key combination to the active window. Click into the **Keystroke** field and press the keys you want to send (e.g. press Ctrl+Enter); the field shows the platform-native symbols (⌘⇧⌥⌃ on macOS, Ctrl+Shift+Alt elsewhere) so you don’t have to translate between platforms. * **AHK Script** – runs an AutoHotkey v2 script file * **AHK Inline** – runs a short AutoHotkey v2 snippet directly The Edit Voice Command dialog includes a **context-sensitive cheat sheet** below the Action field that updates with the Type dropdown – it explains the press-to-capture editor for Keystroke commands, lists common AHK patterns for AutoHotkey Code, names the available internal actions, etc. So you don’t need to memorise the full reference up front. Note **Keystroke commands are cross-platform.** A command captured as **Ctrl+S** on Windows is stored in Qt’s portable format and fires **⌘S** automatically on macOS – the macOS dispatcher swaps Ctrl ↔ Cmd internally to match what every Mac app does. So you can build your command list once and it works on whichever machine Supervertaler is running on. Under the hood, Windows uses `SendInput` (compatible with WPF apps like Trados Studio) and macOS uses AppleScript via `osascript`. ### Removing a command Select the row and click **Remove**, or select multiple rows and remove them together. ### Edits take effect immediately under Vosk When the Always-On engine is **Vosk**, adding / editing / removing / disabling a command immediately rebuilds Vosk’s recogniser grammar in the background – you don’t have to stop and restart Always-On to “teach” Vosk a new phrase. The status bar briefly shows `🔄 Vosk grammar refreshed (N phrases)` to confirm the swap took effect. The next utterance you speak will use the new grammar. *** ## Settings ### Always-On engine The dropdown in the Always-On section picks which speech-recognition backend listens for commands. | Engine | Best for | Speed | Cost | Internet | | --------------------------------- | ------------------------------------------------------------------- | ----------------- | ------------------------ | ------------- | | **Vosk** *(default, recommended)* | Commands only – your phrase list, ignores everything else | Instant (\~30 ms) | Free | No | | **faster-whisper** | Commands + dictation of running text from one continuous mic stream | \~1–3 s | Free | No | | **OpenAI Whisper API** | Same as faster-whisper but offloaded to OpenAI’s servers | \~0.5–2 s | $0.006 / minute of audio | Yes (API key) | **Vosk** is the default for new installs. It’s purpose-built for fixed-vocabulary command recognition: pass it your active phrase list, and it biases the recogniser toward those phrases while silently dropping anything else as `[unk]`. That makes it both faster *and* more accurate for commands than any Whisper variant – and you can leave Always-On running all day for $0 in API fees and near-zero CPU load. **faster-whisper** runs the same Whisper models OpenAI ships, but on a CTranslate2 C++ engine – roughly 4× faster than the original `openai-whisper` package on CPU, with much lower RAM. Choose this if you want **continuous dictation of running text** in always-on mode (every utterance gets transcribed in full, then either typed if it doesn’t match a command, or fires the matched command). **OpenAI Whisper API** sends each utterance to OpenAI’s hosted `whisper-1` model. Slightly faster end-to-end than running faster-whisper locally on most laptops, but each minute of audio costs about $0.006 – so leaving it on all day adds up. Requires an OpenAI API key in **Settings → AI Settings**. The first time you start Always-On with Vosk, the small English model (\~40 MB) auto-downloads to `/vosk-models/`. Same for the small Dutch model when your project’s target language is Dutch. Models are cached forever after the first download. ### Push-to-talk dictation engine The **Dictation engine** dropdown in the Push-to-Talk Mode group controls what handles your push-to-talk dictation hotkey (**Ctrl+Shift+Space** by default). This is independent of the Always-On engine, because the two paths have different needs: | Setting | What runs when you trigger push-to-talk dictation | | ------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- | | **Same as Always-On** *(default)* | Auto-routes: Vosk or faster-whisper Always-On → faster-whisper push-to-talk; OpenAI API Always-On → OpenAI API push-to-talk | | **faster-whisper (offline)** | Always faster-whisper, regardless of Always-On engine | | **OpenAI Whisper API (online, fast)** | Always the API, regardless of Always-On engine. Useful pairing: Vosk for free continuous commands + OpenAI API for fast running-text dictation. | The “ℹ️ Push-to-talk will use: …” indicator below the dropdown shows the *resolved* engine (after auto-routing) so you always know which backend will run. **Why isn’t Vosk an option here?** Vosk’s grammar mode is built for fixed phrases, not free-form transcription. Pressing Ctrl+Shift+Space produces running text, which Whisper handles vastly better. So push-to-talk silently falls through to a Whisper engine even when Always-On is set to Vosk. ### faster-whisper model The Whisper model size dropdown applies whenever a Whisper engine is active – that’s faster-whisper for either Always-On or push-to-talk, *or* the OpenAI API. (The API ignores this setting and always uses `whisper-1` server-side.) Larger models are more accurate but slower and need more RAM. | Model | Download size | Notes | | ------ | ------------- | -------------------------- | | tiny | \~75 MB | Very fast, lowest accuracy | | base | \~142 MB | Good balance (recommended) | | small | \~466 MB | Noticeably better accuracy | | medium | \~1.5 GB | High accuracy | | large | \~2.9 GB | Best accuracy, slow on CPU | ### Mic sensitivity Controls the amplitude threshold used to detect speech onset. * **Low (noisy)** – raises the threshold; ignores quiet background sounds but may miss soft speech * **Medium (normal)** – default; works well in a typical home office * **High (quiet)** – lowers the threshold; captures quiet voices but may trigger on background noise ### Listen for commands only *Whisper engines only.* The checkbox is hidden when the Always-On engine is **Vosk**, because Vosk’s grammar mode already drops non-command speech at the recogniser level – the setting would be a structural no-op there. For **faster-whisper** and the **OpenAI Whisper API**: when checked, Always-On fires voice commands but discards any speech that doesn’t match a command – it is not typed. Use this if you only want voice control with a Whisper engine, not dictation. When unchecked, unmatched speech is transcribed and typed at the cursor position. ### Maximum recording duration Sets the upper limit (in seconds) for a single voice clip. Speech that exceeds this length is cut and transcribed up to the limit. Useful to prevent long silences from being held open indefinitely. ### Language * **Auto** – uses the project’s target language as the transcription hint * Explicit language – forces Whisper to transcribe in the selected language, which improves accuracy when the target language differs from the source *** ## AutoHotkey integration AutoHotkey v2 must be installed for AHK-type commands to work. Supervertaler checks for it automatically and shows the path in the AutoHotkey section of the Voice settings panel. To verify: the status line shows either the AHK path (green) or “AutoHotkey v2 not found” (orange). Click **Open scripts folder** to open the folder where standalone AHK script files are stored. *** ## Using Voice with Trados Studio Voice sends input at the Win32 hardware-input level (equivalent to physical keystrokes), which is fully compatible with Trados Studio’s WPF editor. Useful commands to add: | Phrase | Type | Action | | ------------------ | --------- | ------------ | | ”confirm segment” | Keystroke | `ctrl+enter` | | ”next segment” | Keystroke | `alt+down` | | ”previous segment” | Keystroke | `alt+up` | | ”go to top” | Keystroke | `ctrl+home` | | ”undo” | Keystroke | `ctrl+z` | After creating a command, start Always-On, click into Trados Studio, and speak the phrase. *** ## Global hotkeys | Shortcut | Action | | --------------------------------------- | ---------------------------------------------------- | | **Ctrl+Alt+O** (⌘⌥O on macOS) | Toggle Always-On listening | | **Ctrl+Shift+Space** (⌘⇧Space on macOS) | Push-to-talk (one utterance) – default, configurable | Global hotkeys work on macOS too (via the NSEvent monitor), but require Accessibility permission for whichever binary launched Python – see [Keyboard Shortcuts](/workbench/settings/shortcuts/#per-platform-notes) for setup. All hotkeys can be customised in **Settings → Keyboard Shortcuts**. Note **Always-On moved from Ctrl+Alt+A to Ctrl+Alt+O in v1.10.368.** Supervertaler for Trados now uses Ctrl+Alt+A for “Add term with abbreviation”, and because Always-On is a *global* hotkey it fires whichever application is in front – so a single press in Trados would have triggered both. If you had customised it to Ctrl+Alt+A yourself, it has been returned to the new default; you can set it back, but it will keep clashing while both products are running. *** ## Related pages * [Clipboard Manager](/workbench/clipboard/overview/) * [Keyboard Shortcuts](/workbench/settings/shortcuts/) * [General Settings](/workbench/settings/general/)