Everything Contextra does, in the order you'll use it

Open a document with its source language set and Contextra fills every empty paragraph or cell your memory already knows, with no AI, and you can undo it.

New documents start on Auto-Detect unless your usual language pair or a project's pair gives a source language, and nothing is filled until the source is set. It needs Use memory on, which it is by default. Only empty rows are filled, and a match fills a whole paragraph or cell.

The language chip in the toolbar opens the document's language pair and AI model in one click, with every other document setting one link away. If the source language isn't set, the chip turns amber and tells you why your memory can't match yet.

It's locked while the document is busy. On AI Credits, it lists only the models offered on AI Credits.

From your approved work, context matches (101–102%) apply themselves and a 100% match waits for your review, never overwritten by the AI. The AI drafts the rest.

Approved work means rows you accepted in Contextra, TMX imports and team memories you subscribe to.

If a stored sentence differs from the current one only in numbers or reference codes, Contextra can swap the new figures into your stored translation. You get it as a 100% match to review, and the TM Matches panel shows which figures changed.

Only when the figures are written the same way in both languages; otherwise you get a fuzzy match. The rest of the sentence must match exactly, capitals included.

A sentence that differs only in capitalization is flagged at 99% with the difference highlighted, never passed off as a 100% match.

The highlight is in the TM Matches panel.

Text reused from your memory takes on the current sentence's formatting. When only some words are bold, italic or linked, the result is approximate, so check those rows.

A segment that is one link from end to end loses the link. During Translate, a memory match with mixed formatting can get a small, billed formatting fix from the AI.

When a row has no exact memory match, Translate gives the AI a near match (70% or more) as a reference, for consistent terminology. Your glossary takes priority.

It's guidance, and its text goes to the AI provider with the request.

Memory is on by default. Switching Use memory off in the Library's batch bar stops pre-fill and stops Translate from consulting your memories and Term Bases, while rows already filled keep their text.

The switch isn't in the editor, and it's back on at the next start. Analyze still consults your Term Base, and the TM Matches panel still shows matches.

Memories are matched by language pair: a memory whose language has no region, such as one from a region-less TMX file, serves every regional variant of that language. Memories for Portuguese (Brazil) and Portuguese (Portugal) are never consulted for each other's documents.

Source variants are kept apart too: English (US) isn't English (UK), and each Spanish variant is separate.

Filter the grid to just the rows your memory filled that you haven't accepted yet, or to just the 100% matches still waiting for review.

Two clicks, from the badge legend in the footer. There's no sort by match percentage.

Your memory isn't tied to a file type. A paragraph you approve in a Word document is matched when the same text appears in a spreadsheet cell or on a slide, and the other way round.

Matching is by whole segment, and context matches (101–102%) rarely carry from one format to another.

Once the document's source language is set, each row you accept with Enter is saved to your default memory for that language pair, with the paragraphs around it. The same passage can then come back later as a 101–102% match.

Accepting the same row again replaces what was saved. On the Team plan, keep a memory of your own for each language pair, and your accepts go there.

Contextra never saves its own AI drafts to your memory. Entries come only from rows you accept, entries you add or import yourself, and team memories you choose to subscribe to.

Accepting an AI draft makes it your approved translation, so accept what you've checked. TMX imports are treated as approved work, whoever produced them.

Select a row and TM Matches lists up to 20 matches of 70% or more from every memory for the pair, best first, highlighting how each near match differs from the current source. The first nine are numbered, and ⌘1–9 (Ctrl+1–9) applies one.

At most five fuzzy matches per memory. The numbers work while the TM Matches tab is showing; an AI suggestion, when there is one, takes number 1. The ✦ Generate AI suggestion button under the list is an AI request.

Select any row and the TM Matches panel says how its current translation got there, from the memory tier or the AI model, and who accepted it and when.

It names the tier, not the memory the match came from.

Not sure the stored translation fits? Ask the AI for an alternative right beside the memory match, and pick the one you prefer.

An AI request, counted under AI suggestions in Usage. The AI doesn't see the memory match when it writes its version, and the suggestion is gone when you move to another row.

Select a phrase and press ⌘⇧K (Ctrl+Shift+K) to see where it appears in your Term Bases, your memories and your other saved documents in the same language pair, archived ones included. Hits are capped, and a document hit shows whatever translation was saved there, reviewed or not.

A plain text search of 2 to 120 characters (from 1 in Chinese, Japanese and Korean), with up to 30 Term Base, 50 memory and 50 document hits. It leaves out the open document and Recently deleted.

Set the document's source language, and each glossary term you add or correct is saved to your default Term Base for that language pair. In later documents of the pair, Analyze adds the terms that occur in them to the glossary, and Translate gives them to the AI.

Untouched Analyze proposals are saved only with Save to Term Base; correcting a term's source leaves the old entry. Opening a document adds no Term Base terms: only Analyze and Translate, both AI steps, use them.

Keep as many memories and Term Bases as you like, per client or field. Every one for the document's language pair is consulted, and each pair has one default memory that your accepts are saved to.

Separate memories don't keep clients apart: all of them are consulted for the pair. A memory's languages are fixed once it has entries.

Each memory entry shows when it was added and whether it was accepted, naming the document while it's still in your Library, or imported, and you can correct its text in place. Search every memory or every Term Base at once, with accents and capitals ignored.

The list shows your 200 most recently used entries, and search finds the rest, up to 100 hits. Deleting an entry takes one click and can't be undone.

Import Translation Memories as TMX and Term Bases as TBX (TBX-Basic or TBX v3), and export them as TMX and TBX.

Imports bring plain text, without inline tags or sentence context, so imported memories match up to 100%. An import can replace entries with the same source.

Pre-fill, the TM Matches panel, Concordance, memory search and TMX/TBX import and export never call the AI, so they work without an AI key and spend no credits.

You still sign in to your Contextra account. Adding or editing a glossary term does use the AI (see Glossary).

Your memories and Term Bases are stored on your computer.

AI requests carry near matches and Term Base terms to the AI provider, and sharing a memory with your team uploads it. Memories aren't available while you're signed out.

One click on Analyze has the AI read your document and work out its field, jurisdiction and register, so the whole document is translated with that context.

Optional: you can translate without it, and it runs by itself only in the Library's Analyze & translate. The register isn't displayed, so steer it with your instructions. Comments, charts and slide masters aren't read.

While Analyze runs, you see which of its three steps it's on: reading the document, understanding the context, learning the terminology. Library rows then show the field it detected, and a document with nothing translated yet puts Analyze forward as the next step.

The field is shown unless it's General, and in English in every interface language. Analyze can't be stopped from the editor.

Leave the source on Auto-Detect and Analyze picks the source language from Contextra's list of 27 language variants, with its best guess at the regional variant, such as Brazilian or European Portuguese. It sets it on the document, and you can change it from the language chip.

Only when the source is on Auto-Detect. Check the guess: memories for another variant of the language aren't consulted. If the answer isn't on the list, the document stays on Auto-Detect and no memory or Term Base is used, so set the source yourself.

The AI is told to write its notes to you in your interface language (English, Portuguese or Spanish), not the language you're translating into: term meanings, reasons for skipping and term analyses.

Portuguese means Brazilian Portuguese. Notes written before you switch keep their language, and the note about a Term Base conflict is in English.

Analyze links each glossary term to the segments where it appears, plural and other inflected forms included, so you can jump to them, filter by them and check them. Advanced settings let you tune how it maps terms to segments.

If linking fails, the document has no term links; without the AI, a term you add is linked by its literal text.

Analyze asks the AI to propose glossary entries only for terms that need pinning down: ones translators could render differently, or that recur and must stay the same. It also adds the Term Base terms found in the document, and an empty glossary is a valid result.

Proposed terms come back clean, with any explanation moved to the note. The AI can miss terms or propose too many, and running Analyze again replaces the glossary; terms you added come back only from your Term Base.

When you analyze a document, the Term Base terms found in it take precedence over the AI's proposals for the same terms. If the AI still proposes a different translation for one, the Considered & skipped list shows both.

A term counts as present when it appears as written or as a simple plural, accents ignored; up to 200 are given to the AI.

Right after Analyze, a Considered & skipped list shows the terms the AI reported leaving out, with its reasons, and any proposal that differed from your Term Base. Any of them can be opened in Analyze term and added.

It lists only the skips the AI chose to report, and it isn't saved with the document. Opening a term in Analyze term is an AI request.

The glossary sits beside the grid with every term, its agreed translation, its note and where it came from (your Term Base, this document, or you), and brings the current segment's terms to the top.

The label shows where a term came from, not whether you've checked it: an Analyze term you edit keeps its Doc label. "In this segment" follows the term links, not a text search.

On the segment you're working on, glossary terms are highlighted in the source, and their agreed translations, or listed alternatives, where they appear in the translation. The filter button beside a term filters the grid to the segments linked to it and highlights the term in each row where it can be found.

Only the current row is highlighted, or every row while a term filter is on. A different rendering, or an irregular or contracted form, gets no mark. The filter button shows when you point at or tab to a term; clicking the term itself opens its details.

If the segment you're on doesn't use a linked term's agreed translation, the term's card in the glossary tells you.

A literal text check with capitals ignored, so a correct inflected form or a listed alternative can raise it. Current row only, with the glossary column open.

Click a term to see its meaning and note, how many segments it's linked to, and jump straight to the first one.

To see every linked segment, use the term filter.

Analyze term, an on-demand AI request, gives you a definition, alternative translations labeled with register and when each fits, AI-written usage examples and, when it finds them, warnings such as false friends or jurisdictional gaps. One click on Use sets the one you choose.

All of it is written by the AI, usage examples and their sources included, and none of it is verified: check it before relying on it. The result lasts for the session.

Drag-select a word or phrase inside a source cell to analyze it on the spot and, if it deserves it, add it to the glossary.

An AI request. It works on a mouse selection inside one source cell.

Add a term by hand and Contextra links it to the segments that contain it, then, with a license and AI access, translates those not yet confirmed, untranslated ones included. Rows you accepted, typed yourself, or took from a 101–102% or human 100% match are left alone.

Linking a new or changed term is itself an AI request on your key or your AI Credits, and it runs even while your license is paused. With neither set up, the term is linked by its literal text. Nothing is re-translated while the document is busy.

Remove a term from a document's glossary with a two-click delete right in the list, and your Term Base copy stays.

Because the Term Base copy stays, the term can return at the next Analyze: delete it in Memory ▸ Term Base to stop that.

Change a term's translation and, when a license and AI access are available, the unconfirmed rows linked to it are re-translated. Rows you accepted, took from a 101–102% or human 100% match, or typed yourself are left as they are, counted, and handed back for review.

Linked rows only; a 100% fill from a memory entry that isn't approved work is re-translated too. It's skipped while the document is busy or the license is paused, and rows aren't marked as outdated.

Fix a term in one document and your other open documents in the same language pair that contain it get the corrected term in their glossary. Documents of a compatible field also have their unconfirmed rows re-translated when the AI is available.

Open documents only, and it replaces the other document's own entry for that term. Closed documents get it at their next Analyze or Translate. Not from documents on Auto-Detect.

Save a document's whole glossary to your Term Base in one click, terms Analyze proposed that you haven't edited included, and a notice tells you which Term Base it went to.

It saves at once, with no question first, and unreviewed proposals then count as established terminology in later documents, so check them before you save.

Translate asks the AI to use your glossary's agreed translations, including where Analyze found a term in an inflected form.

With Use memory on, Term Base terms found in the text are included too.

Brief the AI at three levels: general instructions for all your documents (optionally per target language), a workspace project's instructions, and one document's. When they conflict, the AI is told the more specific level wins.

Up to 4,000 characters per box, stored with your account on this computer. They go to Analyze and Translate, not to Analyze term, and team projects have no project level.

Give each client project standing instructions for the AI, such as register, spelling variant and names to keep. They apply to every document in the project whenever it's analyzed or translated.

Rows already translated aren't redone by themselves; Re-translate AI rows is another AI run.

Instructions shape the style (register, spelling, names, titles), and the AI is told that your glossary and the number and date rules take precedence over them.

The order comes from the prompt; the app doesn't check it.

Write general instructions for every target language, a whole language, or just one variant, so a rule meant for UK English never reaches a US English job.

Project and document instructions apply to every target language. A rule for Spanish (Latin America) doesn't reach Spanish (Mexico).

The main instruction boxes (this document, a project page, and the all-languages box in Settings) offer one-click starter lines, and every instruction box saves as you type.

The per-language boxes and the editor's project box have no starter lines, and boxes are locked while the document is busy.

See your instructions exactly as the AI receives them.

The instructions block only, not the whole request, in the editor's Instructions tab.

Contextra warns you when an instruction mentions a glossary term without its agreed translation, or seems to set date or number formats. It also warns when one asks for British or American English, or European or Brazilian Portuguese, against the document's target.

Keyword checks in English, Portuguese and Spanish; the variant check covers English and Portuguese only.

When a document has instructions, its Instructions tab counts the AI rows translated under different instructions and, on request, re-translates only those.

An AI run, per open document.

Choose the number and date format the AI is told to use, or leave it on Automatic for the target language's usual convention. The AI is also told to leave reference numbers, IDs and codes as they are.

An instruction, not a converter: the Quality check compares the values afterwards. Rows filled from memory aren't reformatted.

Describe your house style for dates in your own words, such as month in full, no leading zero or ordinal first day. Add it to the chosen format, or let it replace the format.

The AI follows it through the prompt; nothing rewrites dates by rule.

One thinking-level setting (Low, Medium, High or Dynamic) works across Claude, GPT and Gemini. You can also tune the pacing and, for the models that accept it, a temperature, or keep the defaults.

One setting for all documents. High and Dynamic cost more and take longer, and Dynamic behaves differently per model. Only the Gemini models use temperature.

Press Enter to accept a translated segment: once the source language is set, it's saved to your memory. Then Contextra moves you to the next translated segment you haven't accepted yet.

One row at a time, and Enter does nothing on an untranslated row.

Contextra tracks reviewed separately from accepted: open a translation and move on and it counts as reviewed, without writing anything to your memory.

The cell must be opened and left; moving with ⌘↓ (Ctrl+↓) or scrolling doesn't count. The reviewed total also includes accepted rows and 101–102% rows.

Accept a segment once and Contextra fills and accepts every later occurrence of it in the document, with no AI request. In your other open documents with the same language pair it does the same for occurrences you haven't accepted or written yourself, and Undo takes it back, one document at a time.

An occurrence has the same text, capitals and formatting. In the same document it goes forward only and replaces later occurrences even if they were accepted or typed differently. In other documents: open tabs only, never while translating, and the copies aren't saved to memory with their own context.

Change a confirmed translation and its ✓ or match badge comes off until you accept it again, so edited text never keeps an old approval.

Every way text changes counts: typing, Replace, a source correction, a history restore or Copy source.

Memory matches, AI drafts, accepted and failed rows each carry a badge: 102/101/100%, fuzzy, AI (AI·low when the model reported low confidence), ✓ or ⚠. Text you typed yourself is never labeled as AI.

Confidence is the model's own report, and only low gets its own badge.

The footer shows how far the document is, with a color bar by badge type and translated and reviewed counts. One click opens a plain-words legend of the badges; click one to filter the grid to it.

Reviewed also counts accepted and 101–102% rows. Blank and numbers-only rows aren't counted, and AI·low is counted as AI.

The footer always shows where a document stands (Analyze, Translate, Review, Export), and when the next step is Analyze, Translate or Export, that toolbar button is the highlighted one.

At Review, no button is highlighted: reviewing is yours to do. Analyze is optional.

Reviewing, editing, accepting into your memory, merging and find-and-replace run on your computer with no AI requests. The AI runs only for Analyze, Translate, an AI suggestion or a term analysis, and, when you add or change a glossary term, to link it to its segments and re-translate the rows affected.

Linking a term is an AI request on your key or credits, even while your license is paused. On the Team plan, entries you accept into a memory you share sync through Contextra's servers.

While a document translates you keep working, and rows fill in as they're done, with a live count. When the run ends on the document you're viewing, a summary says what's left and can filter the grid to the failed or AI rows.

No summary after Stop, or when another tab is active at the end. An accept made while the document is still translating can be overwritten by the run.

If a segment that had no translation fails to translate, its cell says so, with the reason when the AI service gave one, and offers a one-click Retry.

A row that already had text shows only ⚠, with no reason or Retry. Reasons are in English, and Retry is an AI request.

Re-translate exactly the rows you select, even ones you already accepted, with one button or ⌘⇧↵ (Ctrl+Shift+Enter).

The result isn't saved to your memory until you accept it. If it fails on an accepted row, the old text stays and its ✓ becomes ⚠.

Open several documents in tabs: each translates and tracks its progress on its own, so you can review one while another runs. Switch with ⌘⇧] and ⌘⇧[ (Ctrl+Shift+] and Ctrl+Shift+[), or jump to one with ⌘⇧1–9 (Ctrl+Shift+1–9).

Each run uses the AI on your key or credits, within your provider's limits on your own key or the AI Credits limits otherwise. Switching to a tab brings in newer memory matches and Term Base terms with no AI request. On some Windows keyboard layouts the bracket keys need AltGr; Ctrl+Tab works too.

The grid draws only the rows on screen, however long the document.

The Word preview draws the whole document, and the Excel preview shows up to 1,000 rows by 60 columns per sheet.

Click a translation to edit it in place, with no pop-up box, and the text doesn't move when the cell opens.

Bold, italic, underline, superscript and subscript are one click away, the first three also by shortcut, while font, size and color always come from the source segment's main style. Paste from Word or the web and only text and those five formats come through, non-breaking spaces included, so pasting brings no foreign font or web link into the client's file.

Editing applies the source's main font to the whole segment, evening out mixed fonts, sizes or colors, and there are no font, size, color or strikethrough controls. Links copied from another Contextra cell are kept; the source cell takes plain text only.

Shift+Enter adds a line break inside a segment, and it arrives as a real line break in Word, Excel and PowerPoint; Excel cells even switch on Wrap text so it shows.

In the translation cell only. Row heights aren't changed.

In Word documents, hover a link to see where it points, and add, change or remove a link right in the translation with ⌘L (Ctrl+L). Export writes every link in your translation back with the right address, and a link you add gets Word's own link style.

Word only: in Excel and PowerPoint a notice explains why not. http, https and mailto links; internal links such as a table of contents are read-only. A link missing from an AI draft isn't flagged, and link text the AI drafts can lose Word's link look, so check linked rows.

The translation you're editing is spell-checked with your computer's own dictionaries: on a Mac in any language macOS has a dictionary for, on Windows when you translate into the language Windows is displayed in. On a Mac, autocorrect, smart quotes and dashes, and text replacement are switched off in the app, so macOS doesn't change what you type.

No built-in dictionaries or grammar check. Switch it off in Settings ▸ Appearance.

⌘⇧C (Ctrl+Shift+C) copies the source into the translation, for one segment or a whole selection, for names, codes and anything that stays as it is. ⇧F3 (Shift+F3) switches the selected words between UPPER, lower and Title Case without losing their formatting.

Title Case capitalizes every word and doesn't follow language rules such as the Turkish dotted i, and many Mac keyboards need fn for F3.

Spot a typo in the source? Fix it in place with ⌘⇧E (Ctrl+Shift+E) so the mistake never enters your memory, while the delivered file keeps the client's original wording and one click restores it.

The next time you accept that row, the entry it saved under the misspelled source in your default memory is corrected too. Team memories you subscribe to are read-only, and pasting into the source gives plain text.

Merge segments with ⌘J (Ctrl+J), even rows that aren't next to each other or a sentence split across table cells, and place the ¶ tags in your translation. Split one with ⌘T (Ctrl+T), and on export each part goes back into its own paragraph or cell, with no cell, row or paragraph added or removed.

Headers, footers, notes and text boxes can't be merged; merging is off while the grid is sorted or translating, and merged rows can't be split. If a ¶ tag goes missing, the ¶! badge and the Quality check flag it, but Export anyway still writes the file.

Every row keeps a fixed number in document order, 12a and 12b for split halves, so numbers never shift when you filter or sort.

They count translatable paragraphs, with headers, footers, notes and text boxes after the body, so they don't match Word's own numbering.

⌘F (Ctrl+F) searches the source, the translation or both, highlights every hit and says how many rows match. Sort by length or alphabetically to sweep similar rows together; the document itself is never reordered.

A plain text search that ignores capitals but not accents, with no regular expressions. There's no sort by match rate, and merging is off while sorted.

Filter the grid by where translations came from, by state (untranslated, not reviewed, needs attention) or by exact match tier, with a live count of what's showing.

Per document, and not saved. The From memory filter drops rows you've since accepted.

Select many segments with shift-click, ⌘-click (Ctrl-click) or ⌘⇧↑ ⌘⇧↓ (Ctrl+Shift+↑ Ctrl+Shift+↓), then merge them, copy their source, re-translate them or limit a replace to them.

Accepting still acts on one row at a time, and re-translating is an AI request.

Find and replace, ⌘⇧F (Ctrl+H), works on translations only, and skips accepted and 101–102% rows unless you include them. Text inside links and ¶ tags is protected, and every changed row goes back up for review.

Literal text only, with no regular expressions, and never in the source. In one document it acts on the rows the grid is showing, so a filter narrows it.

Replace a term across every open document, or every document in the current project, open or not, that translates into the same language. A preview first counts the matches in each and lets you leave any out.

Archived documents and documents that are translating are left out. A document that isn't open is rewritten with no undo.

⌘Z (Ctrl+Z) undoes recent bulk changes in an open document: a Translate run, a Replace all, an accept and what it filled in, a merge or a split. The Undo in the notification works too.

Up to 50 steps per open document, cleared when you close it. Undoing an accept leaves the memory entry, and Analyze can't be undone.

Each segment keeps its last 20 versions (AI draft, memory fill, your edits, replacements), saved with the document and shown as a diff with when it happened, and who for human changes. Restore any of them from the TM Matches panel.

A restore counts as an edit, so an accepted row needs accepting again, and a merge or split starts a new history.

Enter accepts and opens the next segment that needs review, and arrow keys at the edge of an open cell step into the next one. ⌘↑ ⌘↓ (Ctrl+↑ Ctrl+↓) move the selection, and Esc backs out one level at a time.

No key opens the selected row for editing without accepting it: click it, or press Enter to accept.

Press ⌘K (Ctrl+K) to search Contextra's commands, with their shortcuts, and jump to the Library, Memory, Usage, Settings, your own projects or any active document by name.

Archived and recently deleted documents and team projects aren't listed, and neither are Accept or the match keys. Some commands have no shortcut.

A built-in keyboard reference, ⌘/ (Ctrl+/), lists the shortcuts for your platform, such as Ctrl+H for Find and replace on Windows. Nothing is bound to AltGr, and the number shortcuts follow the physical key, so they work on AZERTY.

Shortcuts can't be changed. A few punctuation shortcuts can't be typed on Spanish or German Windows layouts.

Once the source language is set, and between English, Portuguese, Spanish, French, German, Italian and Dutch, the Quality check compares numbers and dates by value, each read in its language's format. It catches 1,500 left as-is in Portuguese, where it reads as one and a half, and a US 03/05/2026 (March 5) copied unchanged, which Portuguese reads as 3 May.

The translation is read in the number and date style set for the document, so a chosen house style isn't flagged. Correct renderings can still be flagged (numbers as words, converted units, 24-hour times, native dates in other languages), and a written-date style on US English can report false day-and-month swaps.

Case numbers, invoice references and ID numbers are compared character by character, so a changed digit, letter, slash or hyphen is flagged as an error. A legal citation re-punctuated for the target language shows as a note for you to judge.

It can't tell a citation from an ID number, and codes with no separator, such as FX123456, aren't checked.

It flags a segment where a linked glossary term's agreed translation doesn't appear, allowing a plural (contrato → contratos) and asking you to confirm a changed ending (advogado → advogada). In a project, it also flags a term whose agreed translation differs in another document's glossary.

Warnings, for linked terms only, and listed alternatives aren't counted. The project check compares glossaries, not the translations themselves.

It spots the same source segment (a paragraph, a table cell, a slide text) translated two different ways in one document.

It compares whole segments, capitals included, so the same sentence inside two different paragraphs isn't compared.

It also flags a missing final full stop, double spaces, unpaired brackets and curly quotes, translations much shorter or longer than their source, a missing ¶ tag, and untranslated rows.

These checks follow Western punctuation: translations into Chinese, Japanese or Hindi, or with German- or Polish-style quotation marks, get warnings you'll need to dismiss.

Open the Quality check tab, or run it from the command palette, ⌘K (Ctrl+K), and click a finding to jump to its row. While the tab is open, the check runs again a moment after each change you commit to a cell, and the tab shows how many errors the last check found.

It checks again when you leave the cell, not as you type, and never in the background. The jump clears a review filter that hid the row, but not a text search or a term filter.

Dismiss a false alarm once and it stays dismissed on that segment, saved with the document. A different kind of finding, or a different number, date, ID or term on the same row, still shows up.

For some checks the kind of problem is what's remembered, so a later problem of that kind on the row stays hidden. There's no list of ignored findings to bring back.

Untranslated rows are summed up in one line, with a link that filters the grid to them.

Failed rows are counted but not shown by that filter.

The Quality check runs on your computer, with no AI, no upload and no extra cost, and it works without an AI key. It works the same on Word, Excel and PowerPoint, with findings in your interface language: English, Brazilian Portuguese or Spanish.

You still sign in to your Contextra account. Numbers-only cells are skipped, and the detail line under a finding keeps English month abbreviations.

Inside a workspace or team project, each document is also compared with the project's other active documents in the same target language. It looks for the same source translated differently, a date in figures in one and in words in another, and a term translated differently in another document's glossary.

If one document can't be read, the cross-document findings are dropped for that run. In a team project, only the copies on your computer are compared.

One click on a project page compares all its active documents, open or not, and shows how many disagreements each has, with a button that opens that document on its findings.

It needs two or more documents, and a document alone in its target language reads as consistent. The total counts each disagreement once for every document involved.

When you export from the editor with something left, Contextra first tells you how much: untranslated rows, translated rows nobody has reviewed (by origin) and Quality check errors. Keep working is the default, so pressing Enter doesn't export by accident, and Export anyway is there for a draft.

From the editor only: the Library's Export archive and a team's Complete run no warning and no check. It counts errors, not warnings, and "not reviewed" includes your own edits.

Translate Word, Excel and PowerPoint files (.docx, .xlsx, .pptx) in one workflow and get the same kind of file back, named with the target language and ready to open in Word, Excel or PowerPoint.

No .doc, .xls or .ppt, no macro-enabled files or templates, and no PDF, RTF, ODT, IDML or XLIFF. Password-protected files can't be opened, and other file types dropped on the Library are ignored.

Contextra translates a Word document's body, including table cells and tables nested inside tables, together with its headers, footers, footnotes, endnotes and text boxes, and every row outside the body tells you where it lives. Each translation is written back into its own paragraph.

Not translated: comments, and text inside tracked changes (accept them first), inline content controls and smart tags; that text stays in the source language and moves to the end of its paragraph, without a warning. Fields such as page numbers in a translated header or footer become fixed text.

Logos, seals, signature images, symbols and footnote marks inside a translated Word paragraph stay in the file you deliver.

An image or mark between two pieces of text moves to the end of its paragraph, without a warning.

A text that appears in many cells is normally translated once and fills every cell that shows it, and Contextra never rewrites numbers, real Excel dates or formulas. Table columns are renamed to match their translated headers, as Excel requires.

Not translated: formula results, sheet names, comments, charts, pivot tables, headers and footers, and validation messages. A translated word that a formula looks for can stop that formula matching, and column widths aren't adjusted.

Decks are translated slide by slide, with titles, text boxes, grouped shapes, tables and speaker notes, and slide numbers and dates stay live fields.

Not translated: layouts and masters, such as a layout's footer, SmartArt, charts, comments and embedded objects. Text isn't re-fitted and can overflow its box. A field between two pieces of text moves after the text, and a date field can keep the source language's format.

Every Excel and PowerPoint row, and every Word header, footer, footnote, endnote and text-box row, shows where it lives: Sheet1!B4, Slide 3 · notes, Footer · first page.

Word body rows, table cells included, carry no label.

Cells and paragraphs that hold only digits and dashes, such as 2025 or 10-20, are copied across instead of translated and don't count toward your progress. Numbers with decimal or thousands separators, and dates, still go through translation so they can take the target language's format.

In Word and PowerPoint those cells show ✓. A Word header or footer holding only a page number counts as numbers-only and becomes fixed text. In Excel, numeric cells aren't segments at all.

Export doesn't rebuild your document from scratch: it works on a copy of the original file, writes the translated text into it, and carries every part it had no reason to change across unchanged.

Word's main document part is always saved again, and a few edits are deliberate: Word's link style, link addresses, Excel table columns, Wrap text, and the watermark when your license is inactive.

Export keeps each paragraph's style, list numbering and page setup, and bold, italic, underline, superscript and subscript follow the translated words, placed by the AI or by you.

AI drafts carry emphasis only if the model returns it, and nothing checks it. In paragraphs with links, fields or mixed emphasis, they can drop Word character styles, strikethrough and small caps, and PowerPoint font sizes and links. Memory fills place mixed formatting approximately, and underlines become single.

Export from the toolbar or with ⌘E (Ctrl+E) and choose where to save; the confirmation names the file, with a button to show it in Finder or File Explorer. It adds a note when the file carries a watermark, when a checkpoint went to your team, or when a translation couldn't be placed back into the file, with how many, instead of failing silently.

When something is left, the export warning comes first. The note doesn't say which rows, and it doesn't count moved images or text that was never extracted; the Library's Export archive and a team's Complete don't show it.

Exporting also saves the terms you curated to your Term Base and sends your Contextra account one summary line: file name, languages, page count, token totals and how the AI was paid for. For a team document it also uploads a checkpoint.

The Term Base save is skipped while the source is on Auto-Detect.

A live preview of your document sits beside the grid, and translations appear in it, in place, as you work. It's rendered on your computer, so your document isn't uploaded to be displayed.

A rendering drawn by Contextra, not by Word, Excel or PowerPoint. Some parts never appear in it, such as tracked insertions, content controls and the other header or footer variant, and the rail folds away on narrow windows.

For Word documents, the preview is drawn from your actual .docx, tables, images, headers and footers included, and translations appear in it as you go.

One continuous scroll with the fonts on your computer, not Word's pages, and one header and footer per section.

Spreadsheets preview as sheets, with column letters, row numbers, merged cells and sheet tabs, so you can see which cell you're translating; presentations preview as slides, with each slide's speaker notes underneath.

Up to 1,000 rows by 60 columns per sheet; hidden rows and columns, charts, images, comments and conditional formats aren't drawn, and dates show in ISO form. Slide layout artwork isn't drawn, and charts and SmartArt appear as labeled boxes.

The preview follows your review: the segment you're on is highlighted in the page, and so are rows you select in the grid. Clicking a paragraph, cell or slide text takes you to its row.

Parts the preview doesn't draw can't be highlighted or clicked. Opening a row from the preview and moving on counts as reviewed.

The preview follows you to the right sheet and tells you when a cell is hidden rather than leaving you guessing. Merged and split segments appear divided between, or joined into, their original paragraphs and cells, the way Export will write them.

Hidden rows and columns are never drawn; the notice lists the possible reasons.

The Export view, ⌘⇧P (Ctrl+Shift+P), is drawn from the very file Export is about to write, translated where done and in the source text elsewhere. A Translated / Original toggle flips between the two in one click, and the view opens on the sheet or slide you're working on.

A rendering, so pages differ from Word's, and it doesn't update by itself: use Refresh. When your license is inactive, it shows the watermark on Word and PowerPoint files before you export.

Everything you import is saved to your Library at once, and each change is saved a few seconds after you make it, with the toolbar showing whether it's saved, saving or needs a retry. Text you're typing in a cell is saved when you leave the cell, with a safety save every five minutes.

Saved on this computer only; there's no cloud sync. A crash, or quitting from the Dock or at logout on a Mac, can lose up to about five minutes of typing in an open cell.

Close the window or quit (⌘Q on a Mac, Exit on Windows), and Contextra first saves your open documents, including the cell you were typing in, then quits. If a translation was running, what it had finished is kept.

Quitting from the Dock or at logout on a Mac, and the restart that installs a Mac update, skip this step.

Once you've signed in, later launches open on your Library with the documents you were working on one click away, and reopen whatever you had open last time.

Per account, on each computer. The cards appear once you have two or more documents, four at most.

Drop Word (.docx), Excel (.xlsx) and PowerPoint (.pptx) files onto your Library, or onto a project's page to file them there, and they're saved at once. A single file dropped on the Library opens straight in the editor, and Add documents and ⌘O (Ctrl+O) work too.

Other file types are ignored without a message. Files added outside a project go to Not in a project.

Import a document you've already translated and Contextra recognizes it by its text, even re-saved or renamed, and offers to open the saved one instead of starting over.

It needs the same source text, so a revised version with one changed word counts as new. It asks; it never blocks.

Find any project or document in your Library by name, including work you've archived; accents and word order don't matter.

Names only: to search inside documents, use Concordance.

Group your documents into projects, one per client or case, each with an optional default language pair: files you add from the project's page, or move in before they're translated, start in those languages.

Flat groups, private to you and kept on your computer: no sub-folders, tags, due dates or members, and one project per document. Changing a project's pair doesn't change its existing documents.

Tick a client's documents on the project's page and run Analyze & translate on all of them in one step. Contextra works through them a few at a time, 2 by default and up to 8, showing each document's progress live.

Each document is its own AI run. Analyze selected re-analyzes documents that were already analyzed and replaces their glossaries, and Stop all stops every AI job in the app.

Mark a finished case as done and it moves to your Archive in one piece: out of the way, still searchable, and back in one click.

Archived work stays in Concordance but is left out of batch runs, the consistency check and replace across documents. A document that's busy can't be archived.

Nothing in your Library is deleted at once: deleted work waits 30 days in Recently deleted, and archiving, deleting, moving, restoring and marking a project as done each offer an immediate Undo. Only Delete now and Empty Recently deleted are permanent, and both ask first.

Undo sits in an eight-second notice. The 30-day clean-up runs while the app is open and signed in; after that, only an earlier backup may still hold an item.

When your Library holds 10 or more documents, Contextra offers to archive finished work you haven't opened in 30 days, and nothing is archived until you click.

Only finished work, and never an open document or a team document.

Tick a batch of documents in a list and move them to a project, archive, export or delete them in one step. Archiving and deleting skip documents that are being analyzed or translated, or waiting in a batch, and Contextra says how many it left alone.

A selection stays within one list. Moving isn't blocked while a document is busy, and a document with nothing translated yet takes the project's language pair. Rename documents one at a time from their menu.

Tab to any document and open its menu with Shift+F10, or Tab to ⋯. Press Delete to move it to Recently deleted, and ⌘Z (Ctrl+Z) to undo the last tidy action while its notice shows.

On a Mac, Shift+F10 usually needs fn; Tab to ⋯ works too.

Hand over or put away a whole case as a single .zip with the original files, translated files, working copies and a manifest, then free up your Library in one click.

Translated files come with language codes in their names, for documents with at least one translated segment, and are watermarked if your license is inactive. The archive runs no export warning, and it can't be imported back in one step.

Save a copy of any job as a .contextra file to archive, move or send it, with the original file inside, and reopen it later in Contextra with its glossary and row history.

Set its language pair again after opening. It's unencrypted and contains the client's document, it opens only in Contextra, and it isn't a deliverable.

On every day you use Contextra, it backs up your Translation Memories, Term Bases and document Library on your own computer, and backups are never uploaded to Contextra.

Only on days the app is open and signed in, and the last seven are kept.

Point Contextra at a folder in Dropbox, iCloud Drive or OneDrive, or on a USB drive, and each new backup is also copied there whenever that folder is reachable.

A cloud-synced folder puts your client data in that cloud. A backup made while the folder is out of reach isn't copied later.

Restoring a backup checks the file before touching anything and saves your current state as a Before restore snapshot, so you can switch back, and the last three are kept. Restore from a backup .zip to set up a new computer.

A restore replaces all memories, Term Bases and Library documents, with no single-item restore, so close open documents first.

Use your own Claude, GPT or Gemini API key with your Contextra plan or trial, or skip the key and spend prepaid AI Credits, with no subscription needed. Switch between the two in Settings ▸ AI whenever you like.

Your own key needs a plan or an active trial; AI Credits need a signed-in account and a positive balance. The choice is per account.

The first-run welcome asks how you want to pay for the AI. Paste your Claude, GPT or Gemini key and Contextra checks it with the provider on the spot, or choose AI Credits and skip keys entirely; a link shows where each provider hands out keys.

It opens by itself for a new account with an empty Library on this computer. A key that works doesn't mean the provider account has credit.

Paste a key and Contextra checks it right away with a free request to the provider, so a typo shows up before your first translation.

It proves the key signs in, not that the account has credit, quota or access to a model. Offline, the key is saved without the check.

On your own key, pick from 17 Claude, GPT and Gemini models, for each provider you've added a key for, per document, with a default already chosen for each provider. Switch to AI Credits and the model menu lists the models offered on AI Credits.

The list is the one in 1.7.1 and changes with app releases. Open documents keep their model until you change it.

No API key? Buy a pack of AI Credits and translate with AI models Contextra provides.

Contextra decides which models are offered on AI Credits, and the list can change without an app update. Each account can use up to 2,000 credits a day, with up to 8 requests at a time. Contextra may also pause AI Credits for everyone for a while, for example when a daily safety limit on its spending with AI providers is reached; requests it refuses aren't debited. Your own key isn't affected.

AI Credits come in one-time packs of 100 for $12, 500 for $50, and 2,000 for $160 (USD, taxes excluded), and each AI request is debited by the tokens it actually used.

Bought on contextra.pro. Your AI Credits balance has no expiry date. It belongs to the account that bought it and can't be transferred or shared across a team, so deleting your Contextra account also deletes any unspent credits.

On AI Credits, your balance shows in the top bar of the Library and the other Home pages, one click from topping up. Once you've used enough credits in a period, the Usage page estimates how many pages it will last at that pace.

Refreshed every few minutes and when you come back to the window, not during a run, and not shown in the editor's toolbar.

Run out of credits mid-run and the segments already translated stay in your document, while a banner opens the top-up page in one click or switches you back to your own key.

Batches already sent when the balance ran out are charged but not delivered, and running them again is charged again.

On AI Credits, a request the app resends, after a dropped connection for example, isn't charged twice.

This covers the app resending a request within ten minutes of the AI's answer, not work you run again: Retry, Re-translate and running rows again are new requests and are charged. A resend while the first is still running can end charged but undelivered.

When a provider rejects a request you see its actual explanation, and a wrong key stops the run at once instead of failing every batch.

The provider's own words, in English.

Segments your memory already confirms (101%, 102% and human-approved 100%) aren't sent to the AI for translation, and requests are set up to benefit from the providers' prompt caching.

Rows with only a near match are still translated by the AI. Analyze always reads the whole document, mixed-format 101–102% matches can get a small billed formatting fix, glossary edits run AI in the background, and caching savings depend on the provider.

See what your AI use on this computer costs, Analyze and term analyses included, by day, step, language pair, document and project: estimated at list prices on your own key, in credits on AI Credits. The headline figures say when they're estimates.

This computer only: your other computers and your team aren't included. Own-key figures leave out failed attempts and cache storage, so your provider's invoice is the final amount.

AI requests made on a document are tagged with that document and its project, so you can see an estimate of what each job cost in AI and download the breakdown as a CSV.

The 50 documents with the most use per period are listed, and the rest are grouped. Use from before September 2026 shows as not recorded.

Technical details shows tokens sent, cached and received, an estimate at provider list prices by model and by step, and your last 25 AI requests on this computer. It lists the price of every model in the own-key catalog, even before your first run, so you can compare models.

The cost columns are list-price estimates in US dollars, not what AI Credits charge.

Free 14-day trial — the full app, no credit card needed. AI isn't included: use your own API key or buy AI Credits.

The 14 days start when you create your account. The Contextra app signs in with an email and password only, so it can't sign in to an account created with Google or Microsoft.

Your license is verified on your own computer: once you've signed in, the app opens without reaching Contextra's servers. The license stays valid for up to 14 days after its last check (7 on a Team seat), and it re-checks in the background whenever you're online.

That's every plan, Perpetual included. Everything except the AI keeps working without a connection inside that window; AI steps always need the internet. The first sign-in is online, and a computer clock more than five minutes off counts as an unverified license: the AI pauses and exports carry the notice.

The work on your computer is never locked away: your Library, memories, editing and export keep working. Analyze, Translate and AI suggestions pause, and exports carry an UNLICENSED TRIAL notice, until your license is active again.

The notice sits at the top of a Word document, on every PowerPoint slide and in the printed page header of each Excel sheet, where it replaces any existing header. It also appears when a paid copy can't verify its license, and team features pause while a Team plan is lapsed.

After your trial, a credit pack alone keeps Contextra's AI working in AI Credits mode, with no subscription needed, and your exports stay clean while your balance is positive.

Switch the app to AI Credits first. At a balance of zero or below, the AI pauses and exports carry the notice from the next check.

When the AI can't run because no key is set up, the license isn't active or your AI Credits have run out, a banner at the top says why and points you to the fix. It names the real problem: trial ended, team plan inactive, license inactive, a computer clock that's off and by how much, or too long offline.

A rejected key, a provider outage or a lost connection shows as an error on the document instead.

Every computer signed in to your account shows up in your dashboard, and signing one out there ends its app session the next time it checks in, within about ten minutes while it's online.

An offline computer keeps working until it reconnects, and signing out doesn't erase its local data. Computers with the same name share one entry.

Group a case's documents into a project with a default language pair, then upload a Word document, pick the translator and the due date, and it appears in their app.

Word (.docx) files only, one file and one translator per upload, up to 200 MB; it appears at the translator's next list refresh, with no notification. Owners and managers only, and a pair change applies to new documents.

On a desktop-size screen, see every document's status, due date and translated-versus-reviewed progress, as last reported by the translator's app. Reload the page for the latest.

Status is hidden in windows narrower than 768px, and the due date below 1,024px. There's no manual status, approval or request-changes step.

Every document keeps a timeline: assigned, started, progress saved, handed back, delivered. The workload view shows each translator's open documents, their size in words and what's overdue, so you can see who's free before the next job.

Sizes are the documents' total source words, counted once the translator opens them, not the words left. Members see only their own row.

Invite translators by email as a Member or Manager and see seat usage on the Team page; the owner chooses the number of seats when buying the Team plan in Billing.

Only the owner changes roles, an account belongs to one team, and a pending invitation holds a seat.

Only the assigned translator and the team's owner and managers can download a document; uploads are checked against their checksum before they're published, and the desktop app verifies every file it downloads.

Every member can see document names, assignees and statuses. Team documents are stored on Contextra's servers and aren't end-to-end encrypted.

On a team, Contextra adds a Team section to the app and shows which team you're working for and your role; solo translators never see it. When something's missing, such as a key of your own or a renewal by the owner, it says so before a job fails.

Without a connection, a seat's license stays valid for 7 days after its last check.

Everything assigned to you in your team's active projects is in one Team list, grouped by project with the most urgent first. Each document shows its status and due date, and overdue work is flagged in red.

The list refreshes when you open it, press Refresh, sign in or act on a document; there are no notifications.

If the connection drops while Contextra is open, the Team list keeps showing the last list it loaded, clearly marked, and documents you've already opened keep working from your Library.

After a restart without a connection, the list is empty until you're back online.

Start working opens the document already set to the language pair your manager chose, with your Translation Memory filled in before any AI runs. It also marks the document In progress on your manager's project page, so nobody updates a status by hand.

The pair applies when the manager set one; memory fills need Use memory on and entries for the pair. On a document you've already worked on here, it replaces your local copy with the team's last upload, so export first or resume from your Library.

Progress reports itself while you work: after each Translate run, when you pause reviewing for about 45 seconds, when you close the document and with every upload. Translated and reviewed counts reach your manager's project page while you're online.

Counts only, never document text. Nothing is queued while you're offline, and the web page shows new counts when it's reloaded.

Progress shows how much is translated and, within that, how much has been reviewed: opened and moved past, accepted, or confirmed by a 101–102% match. You still choose what your memory learns, while your manager sees how much has been reviewed.

Reviewed includes 101–102% rows and rows filled by an accept elsewhere in the document, which nobody opened.

Export whenever you like: when you're online, it also sends your team a checkpoint of your working file without marking the job done, and exporting a job you already handed back reopens it.

A checkpoint that fails isn't retried, and saving alone never uploads.

Complete unlocks only when every translatable segment is translated and marked reviewed, and its tooltip tells you how many are still missing.

Reviewed includes rows nobody opened, and Quality check errors don't block it, so run the check before you hand back.

Complete hands the job back with a finished translated .docx, the same file Export would give you, and your manager downloads it from the project page.

Watermarked whenever Export would be. The last Complete wins.

Download what you sent gives you the exact file your manager received. Reopen lets you correct a delivered job: export it, which moves it back to In progress, then hand it back again with Complete.

Reopening replaces the copy in your Library with the team's version.

Reassign a job midway and the next translator picks up the last version the previous translator exported or handed back, with segments and glossary intact.

Only exported or handed-back versions travel; autosave and Save now, ⌘S (Ctrl+S), never upload.

Team documents you've opened on this computer sit in your Library under their team project, next to your own work, with the same batch tools as your own projects.

No renaming, moving or project instructions: the project is managed on the web.

Contextra saves your work on your computer before it uploads anything, so a failed upload leaves your local copy intact. A team document keeps its link to the team after you close it or restart.

Resume from your Library: Start working or Reopen from the Team list replaces your local copy with the team's last uploaded version, so export first.

Check the team-project documents on your computer against each other for the same sentence translated two ways, a date written two ways or a term that drifts. Replace text across every one you have open, after a confirmation that lists each document and its matches.

Only the documents on your computer, in the same target language. Replaced rows go back up for review, and the change reaches the team when you export or complete.

See how much of your own AI use on this computer went to each team project, in your Usage dashboard.

Your own use only; there's no team-wide usage page.

Give the AI instructions for each team document right in the editor.

Team projects have no project-level instructions set by the manager.

Share a Translation Memory or Term Base with your team in one click, everything already in it and not just what you add next. Each teammate who subscribes gets a read-only copy on their own computer, consulted like their own memories for documents in the same language pair and kept in sync every few minutes.

Sharing uploads every entry (source, translation, neighboring sentences and the author's email) to Contextra's servers, with no confirmation step, readable by Contextra unless the team turned on encryption first. You stay the only writer, and copies stay on subscribers' computers after you stop sharing.

Browse every memory your team shares and subscribe to the ones you want consulted for documents in the same language pair.

On a team with encryption on, subscribe on the website: subscribing from the app misses the entries already there.

A memory you subscribe to is copied to your computer and matched locally like your own, and it's there when you're offline. It's a separate, read-only copy, so a sync never overwrites your own memories.

Keep a memory of your own for each language pair, and your accepts go there.

Subscribe to a colleague's memory and, after the next sync, their entries show up in your matches for the same language pair; an in-context 101–102% match can fill your row before any AI runs.

Needs Use memory on, and on a team with encryption on, subscribe on the website. The Memory page doesn't show who wrote each entry.

Set the order in which your subscribed memories are consulted on this computer; a computer that picks up the subscription later starts with the same order.

It ranks within a match tier only, so a better tier still wins, and later changes don't reach computers that already have the subscription.

Shared memories keep themselves up to date in the background, and a subscription made on the website shows up in the app by itself. A memory you share shows how many of its entries haven't reached the server yet.

It syncs at sign-in, every few minutes, after an export and with Sync now, while the app is open and online.

A team owner or manager can turn on end-to-end encryption: from then on, shared-memory entries are encrypted on the sending computer, and Contextra stores them as ciphertext it can't read. Entries shared before then stay unencrypted and readable by Contextra until they change, and memory names, language pairs, owners and entry counts stay readable.

Optional, for the whole team, and it can't be turned off. Key distribution relies on Contextra's member and device list, which the app doesn't yet let you verify. Team documents aren't covered, and decrypted text still goes to the AI in requests.

A new teammate's computer gets the key automatically the next time a teammate who has it syncs, and until then the Team library says it's waiting for access.

A teammate who has the key needs to be online, and entries that arrived while you were waiting come through only when they next change.

Contextra runs on macOS 13 or later, with separate native builds for Apple Silicon and Intel Macs, and on Windows 10 and 11 PCs with Intel or AMD (x64) processors.

There's one build per Mac processor, not a universal one, so pick yours on the Download page.

On a Mac, Contextra is signed with an Apple Developer ID and notarized by Apple, so macOS opens it as software from an identified developer, checked by Apple. You confirm it once on first launch, as with any app downloaded from the internet.

On Windows you get a standard installer, plus an MSI for IT deployment, each published with its SHA-256 checksum.

The Windows installers aren't code-signed, so Windows shows "Windows protected your PC": choose More info, then Run anyway. The checksum proves the file is intact, not who published it.

One download installs the whole workspace: documents, memories, Term Bases and your Library are handled by an engine that runs on your own computer and comes with the app. You sign in with a Contextra account, and the AI runs at the provider you choose or through AI Credits.

On Windows, the WebView2 runtime is downloaded if it's missing.

On a Mac, Contextra tells you when a new version is ready and installs it in one click; if macOS won't let it replace itself, it sends you to the download page. On Windows it tells you and opens the download page, where you install the new version yourself.

Checks run quietly in the background, and a dismissed notice stays gone until the next version. A Mac update needs a restart, which doesn't run the save-before-quit step. A new version can take up to about an hour to be offered.

On macOS, every in-app update is verified against Contextra's signing key before it's installed, and installing a new version keeps your documents, memories and settings where they were.

Once 1.7 has run, your Library's documents and its new backups can't be opened by version 1.6.1 or earlier, because 1.7 converts saved documents to a new format. Don't reinstall an older version.

The first time you sign in, a short welcome of about two minutes sets your interface language, how the AI is paid and your usual language pair, and you can skip any of it.

Once per account on each computer; reopen it from the command palette.

Tell Contextra your usual language pair once, and new documents start in it, unless they belong to a project or a team assignment with its own pair.

Give it a source language rather than Auto-Detect, so your memory can answer from the first open. Per account, on each computer.

Starting from zero? Drop your files in, or try the whole workflow on a built-in fictional sample before you touch a client file.

Translating the sample uses the AI, on your own key or AI Credits, and the sample becomes a normal Library document.

A six-step Getting started checklist follows your first job and ticks itself off as you really do the work. In the editor it docks under the glossary, or folds into a small pill in the status bar, instead of covering the text.

Rows filled from memory don't count as your work. Per account, on each computer.

Three short tips appear once, when they're useful, and you can turn them off. If you already have work in Contextra on a computer, the first-run guide stays out of your way, and ⌘K (Ctrl+K) brings it back.

Judged per computer, so a new computer shows the guide again.

Press Translate before the AI is set up and Contextra walks you through it, with nothing sent, then offers to run it once you're ready.

For a missing key; a wrong key or an empty credit balance shows up when the run starts.

One sidebar takes you everywhere (Library, Memory, Usage, Settings and back to the editor), shrinks to a slim rail while you translate, and keeps your account, plan and Sign out one click away. ⌘⇧D (Ctrl+Shift+D) jumps between Home and your document.

The Team item appears for team members, and Manage account opens the website.

Contextra has a real menu bar (File, Edit, Translate, View, Window, Help) with the main commands and their shortcuts. On a Mac the traffic lights sit in the app's own top bar; on Windows, browser shortcuts such as reload and print are off, so it behaves like a desktop program.

Not every command is in the menu bar. On Windows the browser's right-click menu is off too, while text fields keep cut, copy and paste.

App-wide settings live on one page with tabs (AI, Translation, Advanced, Appearance, Backups, Account), opened with ⌘, (Ctrl+,), and ⌘, or Esc takes you back to where you were.

Document settings live in the editor, in the language chip and the Instructions tab, and project settings on the project's page.

Contextra confirms what it did, with an Undo or a Show in Finder when it helps, without getting in your way. Hover a toolbar or sidebar button to see what it does and its shortcut.

Many disabled buttons don't say why; when the AI is paused, the banner at the top does. Error details from the engine are in English.

Resize the glossary, preview and matches panels to taste and they remember their size; on a small screen the grid keeps its room while the preview steps aside into a tab.

The smallest window is 960 × 600, and sizes are kept per computer.

Use Contextra in English, Brazilian Portuguese or Spanish, or let it follow your computer, and switch at any time without restarting. Menus, buttons, the app's own messages, the preview's labels and the menu bar follow, and dates, numbers and language names are written your language's way.

Portuguese from Portugal gets the Brazilian interface, and other languages get English. Error details from the engine or the AI provider stay in English, and on Windows, Cut, Copy, Paste, Select All and About stay in English in the menu.

Work in light or dark, or let Contextra follow your system as it changes; in dark mode your documents still look like documents, with white Word pages and sheets and slides in their own colors. Contextra also honors your system's reduced-motion setting.

A deck with a dark design previews dark, and speaker notes follow the app's theme.