— A 50-page job, before lunch —
Translate faster, deliver better
No cardFirst 5 pages freeYour work never trains AI
Built for work you answer for
Private by default
Your text is never used to train AI, not by us and not by any provider we use, and it stays out of our logs. Delete your account and your text is deleted with it.
How we handle dataSecurity you can check
Hosted in Germany, in Hetzner’s ISO 27001-certified data center. Every sub-processor is listed, and a DPA is one email away.
All sub-processorsGive it context
Glossaries, styleguides, translation memory and per-string notes go with every request. You set them up once per project.
GlossariesYou’re the human in the loop
The AI drafts; you decide. Review every change side by side, keep or revert it, and send a review link when a client needs to look.
Review linksAPI, MCP and webhooks
The full API on every paid plan. Connect Claude or another AI assistant over MCP, or start runs from your own tools.
Developer guideSomething bigger?
Large catalogs, custom workflows, a DPA for procurement. Transept stays self-serve; write to us for the rest.
support@transept.aiA working day, in five steps
I.Drop in any document, see it translated
Upload a DOCX, EPUB, MD, TXT, or PDF file, or paste plain text. It splits into blocks in a side-by-side editor, and you choose the target language as you import it.
II.A glossary per client that keeps terms consistent
Build it once and switch it on for each project. The client’s terms stay the same from page 1 to page 500, in one language or in twelve.
III.A styleguide that steers every rewrite
Tone, rules, and conventions for each client, generated from a sample page, then versioned and reused. When you re-translate a sentence with a direction ("more formal," "keep the metaphor"), the styleguide steers the rewrite so the result stays in the client's tone.
IV.Literess, your second reader
A literary editor at your side. She reads the translation and flags where the meaning has drifted, the rhythm has gone flat, or the tone has slipped: the small misses that creep in after twenty pages of editing. She also remembers your projects and preferences from one document to the next.
V.Proofread before you deliver
A final QA pass that catches what a tired reader misses: inconsistent terminology, missed glossary terms, slips in style. Run it on a selection or on the whole document.
Show clients your post-editing
When a client asks how much of a translation was machine output and how much was your work, send them the report. It counts your edits by word, lists every paragraph you reviewed, and carries your confirmation that you read the whole text.
- Counted by word. The words the AI wrote, the words you changed and the paragraphs you wrote yourself, each counted on its own.
- Every step on record. Translations, rewrites, fixes, your edits and your approvals in one list, each line marked as yours or the AI's.
- Your sign-off, dated. Before the PDF downloads, Literess asks whether you read the translation through. Say yes and the report records it.
Two reports on the Free plan, unlimited on every paid plan.
Originality report
Chapter 3 · Ukrainian → English
Reviewed in full by the author, confirmed on Oct 21, 2026
Simple pricing
1 credit = 1 source word. Pro mode costs 3 credits per word. What each action costs
Free
Always freeFor your first projects
Unlimited users
Translated and proofread on Standard. About five with Pro.
- Pro mode: the model thinks each passage through, 3 credits per source word
- Standard quality: fast, accurate, faithful
- Translation memory: your past translations, reused to keep new ones consistent
- Glossary, styleguide, QA, import / export: all included
- Notion, Drive, TMX / XLIFF: import, export, and bring your own memory
- 68 languages: from Spanish to Tagalog
- Paragraph-by-paragraph translation in every mode
- Literess, your AI literary editor: 10 messages free, then credits
- Originality reports: 2 included
- No card required
Lite
The pocket edition
Unlimited users
- Whole pages and documents in one run
- Unlimited originality reports
- Workflows: chain translation, proofreading, and checks into one run
- API, MCP & webhooks: automate runs, or let AI assistants work in your account
- Teams: create as many as you need and invite anyone for free
- Top up anytime: pack credits never expire
Starter
Most popularFor regular client work
Unlimited users
- 3× Lite’s credits: for weekly work
- Teams: invite anyone for free; add €29/$33 seats to fund a shared credit pool (80,000 credits a month each)
Pro
For higher volumeFor agencies & large catalogs
Unlimited users
- 3× Starter’s credits: headroom for batch work and full catalogs
- Priority support: a direct line and faster replies
- Predictable bill: most months need no extra credit packs
Nothing could translate what we wrote — so we built what could
Two writers, one problem.
— Vitalii.I wrote 120,000 words of satirical romance fantasy: mages drowning in debt, a multiverse economy that makes Wall Street look ethical, and a love story tangled up in all of it.
— Mariia.I write semi-fictional stories about real people, built around how it felt. Most of the people I wrote about were foreigners. We met in English, spoke in English, lived those moments in English. So I wrote them that way too.
Vitalii VlasiukCo-founder
Mariia IvakhnenkoCo-founder
“But I sometimes wonder what version of them I missed by not writing them in their language.”Mariia
— Vitalii.The same flatness hit me from the other side. Every tool I tried preserved the words but killed the voice. Tone, rhythm, wit, emotional texture: nothing held. It all came back flatter than it went in.
— Both of us.So we did what we know how to do. We read the machine-translation papers. We benchmarked Claude, GPT, Gemini, and a handful of open models on the same 50-page chapter and judged them on accuracy, tone, and whether the satire still landed. None of them held all three; each failed in a different way. So we built the tool we wished had existed when we started.
We still use it every day. This site is localized in Transept: every change to its German, Japanese and Chinese pages goes through the same workflow you'd use.
FAQ
- Upload DOCX, EPUB, MD, TXT, or PDF files, or import pages from Notion and Google Drive. You can export as DOCX or Markdown, or straight to Google Docs or Notion, and an EPUB goes back out as the same e-book. A PDF can be translated as text only or written onto its own pages.
- Standard mode is fast and accurate and uses 1 credit per source word. In Pro mode the model thinks each passage through before it writes, for higher-quality output, and it costs 3 credits per source word. It is usually worth it for literary or nuanced material.
- Nothing already translated is lost, and your balance never goes below zero. If your balance covers only part of a document, you can translate the part it covers and finish the rest after adding a credit pack: Small (30K / €19/$22), Medium (100K / €55/$63) or Large (300K / €149/$170). Pack credits never expire.
- Refunds cover things that went wrong on our side: a technical fault that blocked your work and that we couldn’t fix within 3 business days, or a billing mistake such as a double charge. Email support@transept.ai and tell us which document or payment it concerns and what went wrong. If a single job went wrong, we usually return its credits. A preference for different wording isn’t grounds for a refund, which is why the Free plan gives you 5,500 credits a month to try the quality. EU consumers can also withdraw within 14 days of a purchase and get back the part they haven’t used. You can cancel anytime and keep your plan until the end of the period you’ve paid for.
- Both. Add terms by hand or generate a glossary from existing content, then switch it on or off for each translation.
- Share a document with specific people, comment on any block, and resolve each thread once it is settled. The activity feed shows what changed.
- Teams and email invites are built in: add your translators, share resources, and manage projects from one workspace.
- The originality report counts your edits by word against the machine output, lists every paragraph you reviewed and every fix you accepted or declined, and records the date you confirmed you read the whole text. The Free plan includes two reports, every paid plan has no limit.
4× cheaper model, matched
Give a model the decisions behind past translations, not just their final wording, and it stops guessing. With enough of that context, a model costing a quarter as much matched a premium one on our own benchmarks.
The memory betFeatured on
