Transept

Control

Translation memory that remembers your decisions

Where every translation decision becomes memory — the wording, and the reasoning behind it: the alternatives you rejected, the comments that explained the call. We call it decision-context memory, and Transept feeds it back to the model on the next run. Your decisions from page 1 come back on page 500, on the next document, and on your teammate’s.

Start translating→

Free to begin · No card required

— Don’t take our word for it —

The editor, in miniature

A working slice of the real thing — Literess, glossary, styleguide, workflows, and the translation memory are all live. Click around.

Chill, love

Here to help you translate

Memory
Import
Try:
Search
Scope
AI context
Manage TM settings
Showing matches for the selected block
71%

A key turned in the lock and the door swung open.

У замку повернувся ключ, і двері розчахнулися.

Chill, love — chapter 2you · last week
This doc55%

The knock came just before midnight.

Стукіт пролунав перед самою північчю.

Chill, loveyou · today
Auto-saved · just now
Interactive demo — click anything
In context✦

Every translation tool has a translation memory — Trados and memoQ built the category, and you can get one free in an open-source editor. So the real question isn’t whether a tool remembers; it’s what the memory does. The old answer is retrieval: have we translated this exact segment before? Pull the match, fill the gap. Transept’s answer is context — what does this project already know, and how much of it belongs in the sentence you’re on? Approve a translation and it’s remembered automatically, then surfaced as you work and handed to the model as reference, so the next draft reuses your wording instead of starting cold. No database to build, no concordance to maintain, no copy-pasting from a panel. It’s the difference between a memory you query and a memory that’s in the room while you translate.

Translate it once, reuse it everywhere

Classic CAT tools keep a translation memory you maintain by hand and copy from manually. Transept builds it for you as you work, and — the part generic AI tools can’t do — feeds it back into the model, so the next translation already knows your past decisions.

  1. It remembers as you go

    Every block you approve is added to your memory automatically — your own work and your team’s. No separate database to maintain, no export step. The moment a translation is active, it’s reusable.

  2. Matches surface as you translate

    Select a block and the Memory panel shows the closest past translations — matched by meaning, not just wording. A semantic search surfaces segments that mean the same thing even when they’re phrased differently, fused with exact and fuzzy text matches; a percentage badge shows how close the text is. One click copies the target or drops it into the block.

  3. The AI translates with your memory

    The part a glossary can’t do and a raw LLM won’t do: your matches become reference context in the prompt, so translate, proofread, and Find reuse your established wording instead of inventing a new phrasing each time.

  4. Bring your own, search across everything

    Import an existing memory as TMX or XLIFF and it’s searchable from day one. Scope the search to one document, a project, or the whole team — and search past comments and chat alongside the translations.

Where reuse pays for itself

Repetitive documents

Manuals, contracts, release notes, product catalogs — the same sentence translated once and reused wherever it recurs, instead of re-translated and re-charged.

Consistency across a library

A help center or a doc set where the same warning, the same step, the same boilerplate has to read identically in every article and every language.

Memory that remembers the why

Not just the final wording — the rejected alternatives and the comments behind a choice come back too, so the model reuses your judgment, not only your words.

Where it’s different✦

On matching alone, the field has a ceiling. Trados and memoQ are the state of the art of the old paradigm — exact and fuzzy matches, and at the top, context matches (memoQ’s 101%: a segment plus its neighbors). It’s a database you query, and it’s been perfected. What it can’t do is find the sentence that means the same thing in different words. Transept fuses three kinds of matching — exact, fuzzy, and semantic (meaning-based) — into one ranked result, so a paraphrase surfaces alongside a near-duplicate. Semantic search on its own is risky (it’ll wave through a near-twin like “ileum” for “ilium”); fusing it with exact and fuzzy matching is what keeps it precise.

The bigger gap is where the memory lives. For Phrase, Smartcat, and the rest it’s an organized database beside the editor — you query it, you fill the segment, the model picks an engine. Lilt goes further and fine-tunes a model on your past work, but that’s weights you can’t inspect, not retrieval you can. Transept keeps the memory live: fed into the draft and the QA pass, searchable across your past comments, chats, and whole documents — not only segment pairs — and travelling with the project instead of sitting in a box you remember to open. (For string-level software localization, with a memory keyed to message IDs and wired into CI/CD, a tool like Lokalise is still the better pick — Transept is built for documents and content.)

FAQ✦

Questions, answered without the fluff

  • Yes — it’s built into every plan, including Free. There’s no separate add-on and no extra database to manage. Every translation you approve feeds your memory automatically.
  • A glossary pins individual terms — a name, a product word, a phrase. Translation memory matches whole segments — sentences and paragraphs you’ve already translated. They’re complementary: the glossary keeps words consistent, the memory keeps you from re-translating the same sentence. Most professionals use both.
  • Yes — export it as TMX or XLIFF and upload it. Your existing source / target pairs are searchable immediately and feed the AI like any other memory.
  • No. Words are spent translating new text. Copying, inserting, or reusing a match from your memory costs nothing.
  • Trados and memoQ are the best of classic CAT memory: a segment database with exact, fuzzy, and context (101%) matches. Two differences. First, matching — they stop at text similarity, while Transept adds semantic search, so it finds a sentence that means the same thing even when it’s worded differently. Second, where the memory lives — theirs is a panel you query beside the editor; Transept’s builds itself from the work you approve and is fed into the AI draft and the QA pass.
  • Phrase and Smartcat are strong platforms that organize memory across projects and clients — Phrase even auto-selects an MT engine and scores each segment. But it’s still match-and-fill: the memory is a database the workflow draws from, with the model picking an engine rather than your memory steering the draft. Lilt instead fine-tunes a model on your past work — powerful, but that’s weights rather than retrieval you can inspect or import. Transept’s memory is the context the AI translates inside, built from hybrid retrieval you can see.
  • For document and content work, usually not — translate, memory, glossary, QA, and export live in one place. For string-level software localization (a TM keyed to message IDs, wired into CI/CD), pair Transept with a tool like Lokalise. Import and export run on TMX and XLIFF, so nothing is locked in either way.
  • It’s translation memory adapted to the LLM era. Instead of only retrieving a past segment for you to paste, your matches are passed into the translation prompt as reference, so the machine-translation step reuses your established wording. The same idea as a classic TM, applied to how the AI actually generates the text.
Related✦

Keep reading around this capability

❧

Stop paying twice to translate the same sentence

Start translating→

Free to begin · No card required