Technical document translation that understands the boundary between human language and machine language. Choose your AI agent — from fast and affordable to maximum fidelity. Code blocks, Mermaid diagrams, API references, and brand terms stay untouched. Your docs read naturally in hundreds of languages — and still work on the first copy-paste.
We tried every translation service. They all failed on technical docs.
We maintain 40+ technical documents — API references, architecture guides, infrastructure specs — totaling nearly 200,000 words. We needed them in 11 languages for our global user base.
We tried AWS Translate. It translated our code blocks. It mangled our Mermaid diagrams. It transliterated "The Trinity Beast" into local scripts. It turned GET /admin/health into gibberish. Every translated document was broken on arrival.
We tried manual review. At 40+ docs × 11 languages = 440+ files, that's not a workflow — it's a full-time job.
So we built our own engine. Sentinel preprocessing extracts everything that shouldn't be translated — code blocks, inline code, Mermaid diagrams, brand terms, technical identifiers — replaces them with numbered placeholders, sends only the human-readable text to your choice of AI agent on Amazon Bedrock, then reassembles the document with every technical element intact.
After more than 8,400 successful translations with a 97% success rate, we realized: if we needed this, other teams with technical documentation probably do too.
Now it's a service.
Powered by AI agents on Amazon Bedrock — from Qwen and DeepSeek for high-volume value, to Claude Sonnet and Opus for maximum fidelity. You choose.
Translation engines that understand code don't exist. Until now.
Code blocks, inline code, CLI commands, API paths, query parameters — all extracted before translation and restored after. Your curl commands still work. Your Mermaid diagrams still render. Every CSS class, inline style, HTML attribute, and color value stays exactly where you put it. Copy-paste still works in every language.
Customized brand term protection that ensures your product names, person names, and technical identifiers never get translated or transliterated. They stay exactly as written in every language. No more "三位一体の獣" where "The Trinity Beast" should be.
Powered by AI agents on Amazon Bedrock — your documents can be translated into any of 127 languages, from Spanish and Japanese to Bengali and Swahili, and from any of them into any other, English included, with no pivot through a third language. Submit your documents, pick up to 12 languages per quote, and get them all back — ready to publish.
Not days. Not hours. A few minutes per language for a standard document — even faster with Haiku. Every language in your quote runs on its own worker at the same time, up to 12 at once. Submit and walk away — you'll get an email when it's done.
📚 See It In Action
Our entire 40+-document technical library — API references, architecture guides, infrastructure specs — is translated into 11 languages using this service. Every code block intact. Every diagram untouched. Browse the results yourself.
Same protection on every agent. Leave the choice to Auto, the default, and you get the best agent per language: Qwen 3 235B for every language it supports, and Claude Sonnet 4.6 for Amharic, Hausa, Mongolian, Pashto and Somali, at the same price. Pick an agent yourself and it translates every language in your quote, retries included.
$3.00 per translated document, per language, for any document up to 300 KB — five of our six agents are always $3.00, and Claude Opus starts at $3.00. Choose by quality and specialization, not price.
Excellent quality with specialized strengths. Pick the agent that matches your content and target languages.
A 235-billion parameter mixture-of-experts model with 22 billion active parameters per inference. Handles CJK scripts (Chinese, Japanese, Korean), South Asian languages (Hindi, Urdu), and Arabic with strong natural fluency.
Best for: CJK and South Asian scripts, high-volume batches, iterative re-translations.
A 671-billion parameter MoE model built for complex reasoning. Excels with technical documentation — understands code context, API references, and system architecture deeply. Strong handling of mathematical and scientific terminology.
Best for: Technical documentation, API references, scientific content, code-heavy docs.
The European language specialist. Built by Mistral AI in Paris, this 675-billion parameter model has native-level understanding of French, German, Spanish, Italian, and Portuguese — including idiomatic expressions, regional variants, and formal/informal registers.
Best for: European languages, EU regulatory/compliance content, Romance and Germanic language fluency.
The fastest agent available. Purpose-built for speed while maintaining the instruction-following precision that makes Claude models excellent at preserving code blocks, brand terms, and document structure.
Best for: Fastest turnaround, Latin-script languages, frequent re-translations, straightforward docs.
When nuance matters as much as accuracy. Choose Claude Sonnet 4.6 for your most nuance-heavy content.
Your pick for nuance-heavy content. Sonnet brings deep linguistic understanding to complex scripts and nuanced technical content. It navigates the subtleties of Japanese honorifics, Arabic morphology, and Hindi compound words with confidence — producing translations that read as if a native speaker wrote them, not translated them.
Best for: Complex scripts (Arabic, Hindi, Japanese, Chinese), nuanced technical explanations, content where natural-sounding prose matters as much as accuracy.
When precision is non-negotiable. The only agent where large documents may exceed the $3.00 floor.
The most capable model in the Claude family. Opus brings the deepest reasoning to every sentence. For documents where a subtle mistranslation could cause real problems, Opus delivers the highest confidence output available. Every word is weighed, every nuance preserved.
Best for: Legal contracts, medical documentation, compliance materials, regulatory filings, complex grammar (Finnish, Hungarian, Turkish).
All agents use the same sentinel preprocessing, the same brand term protection, and the same multi-layer validation pipeline. The difference is the depth of linguistic reasoning and the language specialization. Your code, diagrams, and technical identifiers are protected identically regardless of which agent you choose.
Four steps. No manual review. No broken code.
Give us the URLs of your HTML documents — up to 42 in one quote. Choose up to 12 target languages from our 127.
Get an instant, fixed price — $3.00 per document, per language (Claude Opus can be more on larger documents). We show which language each document is written in, and ask only when we're unsure. Approve to start.
Sentinel preprocessing extracts code, diagrams, and brand terms. Your chosen AI agent translates only the human-readable text. Multi-layer validation ensures nothing was lost.
You receive the translated HTML files — ready to deploy wherever you host your docs. Same structure, same styling, same classes and colors, every technical element intact. You get an email when it's done.
Each quote and acceptance has a clear ceiling. Need more? Submit additional quotes — there is no limit on how many quotes you can create.
A single acceptance can process up to 9 quotes of up to 42 documents each — 378 documents, each translated into up to 12 languages. All translations deliver in under 3 minutes per document, per language. One approval, one charge, one consolidated receipt.
See the entire workflow in action — from document submission to translated output.
Walk through both the Dashboard UI and API workflows. See how documents are submitted, quotes are generated, and translations are delivered — with code blocks, diagrams, and brand terms protected throughout.
One request for the quote. One call to accept. Done.
Every translation account gets an API key — visible in your dashboard under Account Settings. Use it for programmatic access (CI/CD pipelines, automated doc builds) or skip the API entirely and submit jobs from the dashboard UI.
Submit all your documents, all your languages, and your chosen agent in a single request. The system analyzes each document and calculates the full order price — $3.00 per document, per language.
POST /translate/quote
Content-Type: application/json
Authorization: Bearer YOUR_API_KEY
{
"docs": [
"https://your-site.com/docs/api-reference.html",
"https://your-site.com/docs/architecture-guide.html",
"https://your-site.com/docs/getting-started.html"
],
"langs": ["es", "fr", "de", "ja", "zh", "pt", "ru", "hi", "ur", "it", "ar"],
"model": "qwen3-235b"
}
// Response — 3 docs × 11 langs = 33 translated documents × $3.00 = $99.00
{
"data": {
"quote_id": "qt-a1b2c3d4-e5f",
"docs": [
{ "doc_url": "https://your-site.com/docs/api-reference.html", "doc_name": "api-reference.html",
"analysis": { "size_bytes": 63488, "estimated_chunks": 4, "difficulty": "moderate", "code_blocks": 12, "diagrams": 2 },
"total_price": 33.00 },
...
],
"doc_count": 3,
"languages": ["es", "fr", "de", "ja", "zh", "pt", "ru", "hi", "ur", "it", "ar"],
"lang_count": 11,
"total_pairs": 33,
"model": "qwen3-235b",
"auto_routed": false,
"model_by_lang": {},
"delivery": { "mode": "express", "label": "Express — real-time translation", ... },
"pricing": { "price_per_pair": 3.00, "total_pairs": 33, "total_price": 99.00, "currency": "USD" },
"expires_at": "2026-10-13T20:00:00Z",
"accept_url": "/translate/accept/qt-a1b2c3d4-e5f",
"sources": [
{ "doc_url": "https://your-site.com/docs/api-reference.html", "doc_name": "api-reference.html",
"source_lang": "en", "source_provenance": "declared", "source_lang_confirm_needed": false, "source_hint": "" },
...
],
"superseded_quote_id": ""
}
}One quote ID, one call. Your payment method is charged and translation starts immediately — results arrive in under 3 minutes per document. No cold starts — the worker is always on.
POST /translate/accept/qt-a1b2c3d4-e5f
Authorization: Bearer YOUR_API_KEY
// Response — every document was fetched again and checked before the charge
{
"status_code": 200,
"data": {
"status": "accepted",
"quote_id": "qt-a1b2c3d4-e5f",
"job_id": "019e4b8a37bc-1f2e3d4c5b6a7980",
"model": "qwen3-235b",
"payment_id": "pi_3abc123def456",
"amount_usd": 99.00,
"status_url": "/translate/status/019e4b8a37bc-1f2e3d4c5b6a7980",
"notice": "", "notice_key": "", "updated_docs": [],
"key": "", "message": "", "refunded": false, "refund_amount": 0
}
}Poll for progress or wait for the email notification.
GET /translate/status/019e4b8a37bc-1f2e3d4c5b6a7980
Authorization: Bearer YOUR_API_KEY
// Response — every pair, the ETA while it runs, the free retry and any refund
{
"status_code": 200,
"data": {
"job_id": "019e4b8a37bc-1f2e3d4c5b6a7980",
"quote_id": "qt-a1b2c3d4-e5f",
"state": "succeeded",
"model": "qwen3-235b", "model_by_lang": {}, "model_display": "Qwen 3 235B",
"docs": ["api-reference.html", "architecture-guide.html", "getting-started.html"],
"langs": ["es", "fr", "de", "ja", "zh", "pt", "ru", "hi", "ur", "it", "ar"],
"total_pairs": 33,
"counts": { "queued": 0, "running": 0, "succeeded": 33, "failed": 0, "cancelled": 0 },
"pairs": [
{ "doc": "api-reference.html", "lang": "ar", "state": "succeeded", "job_id": "019e4b8a37bc-1f2e3d4c5b6a7980", "attempts": 1 },
...
],
"eta": null,
"retry": { "allowed": false, "state": "closed", "retry_job_id": "", "window_ends_at": "" },
"failed_pair_refund": { "status": "", "amount_due": 0, "amount_refunded": 0, "pairs": [], ... },
"downloads_ready": true,
"download_url": "/translate/download/019e4b8a37bc-1f2e3d4c5b6a7980",
"message_key": "tbts_status_succeeded",
"message": "All 33 translations are ready to download.",
...
}
}
// Your files: private links that last 7 days, renewable for 93
GET /translate/download/019e4b8a37bc-1f2e3d4c5b6a7980Side-by-side: what happens to your technical docs with other services vs. ours.
| Feature | Generic Translation | Trinity Beast Engine |
|---|---|---|
| Code blocks preserved | ✗ Translated/corrupted | ✓ Extracted & restored |
| Mermaid diagrams | ✗ Syntax broken | ✓ Untouched |
| Brand terms | ✗ Transliterated | ✓ Customized brand protection |
Inline code (curl, paths) |
✗ Often translated | ✓ Never touched |
| RTL languages (Arabic, Hebrew, Persian, Urdu) | ✗ Layout breaks | ✓ Marked right-to-left automatically |
| HTML structure & attributes | ✗ Tags/attributes corrupted | ✓ All classes, styles, IDs, and attributes preserved |
| Turnaround | Minutes (but broken) | < 3 min per translation (and correct) |
Powered by AI agents on Amazon Bedrock, we translate into and out of 127 languages — every one of them can be your source or your target. Here are some of the most popular:
+ Swahili, Punjabi, Tamil, Telugu, Malay, Czech, Romanian, Hungarian, Swedish, Finnish, Danish, Norwegian, Hebrew, Persian, and many more — 127 in all, including Mexican Spanish, Canadian French, European Portuguese, Traditional Chinese, and seven spoken Arabic dialects alongside Modern Standard Arabic.
Pay per document. No subscriptions. No surprises.
Real-time inference — your job starts immediately and results arrive in minutes. The sentinel preprocessing pipeline protects your code blocks, diagrams, and brand terms while your chosen AI agent translates only the human-readable text.
One price per translated document, regardless of which agent you choose. Your quote shows the exact total before you approve — no hidden fees, no per-token math, no surprises.
| Pairs | Per Pair | Total |
|---|---|---|
| 50 | $3.00 | $150.00 |
| Pairs | Per Pair | Total |
|---|---|---|
| 110 | $3.00 | $330.00 |
Claude Opus 4.8 is the one exception — it is priced on its real cost with a $3.00 floor, which it passes only on documents of about 200 KB or more. Your quote always shows the exact price. All other agents: $3.00 per document, per language — regardless of document size or complexity.
Adjust the sliders to see your total. $3.00 per document, per language with every agent but Claude Opus, for any document up to 300 KB.
$3.00 per document, per language on five of our six agents, for any document up to 300 KB. Claude Opus starts at $3.00 and can cost more on larger documents — your quote always shows the exact price.
If your documentation has code in it, this is for you.
Endpoint references, request/response examples, authentication guides, error code tables. The code stays executable. The explanations read naturally in every language.
Architecture docs, deployment runbooks, CloudFormation templates, Terraform configs. Technical identifiers, resource names, and CLI commands stay untouched.
Setup guides, SDK tutorials, integration walkthroughs. New developers in any country can follow the steps in their language and have working code at the end.
Enter your email to create an account. You'll get a magic link — no password needed. Add a payment method in your dashboard, submit your first document, and see results in 3 minutes or less.
Already have a Trinity Beast account? The Translation panel is already in your dashboard.
100% of revenue from this service goes directly to Cross Power Ministries of Pakistan — funding freedom from brick kiln debt bondage, clean water, medical camps, and education. When you translate your docs with us, you're not just getting better translations. You're changing lives.