AI translation for in-house language teams
Translation that follows your glossary, and shows its work.
Syngram turns your glossaries, past translations and style rules into an AI translation engine. It checks every segment against them, fixes what it missed, and sends only the risky ones to your reviewers.
- 12Clean
The invoice must be paid within 30 days.
Die Rechnung ist innerhalb von 30 Tagen zu bezahlen.
- 13Clean
Acme Cloud provides the services described in this agreement.
Acme Cloud erbringt die in diesem Vertrag beschriebenen Leistungen.
- 14Review
Late payments incur a late fee of 2% per month.
Bei verspäteter Zahlung fällt eine Verzugsgebühr von 2 % pro Monat an.
High risk: payment terms. Sent to a reviewer.
- 15Clean
Either party may terminate this agreement with written notice.
Jede Partei kann diesen Vertrag schriftlich kündigen.
Why Syngram
Generic machine translation doesn't know your terms.
So your language team rereads every line. Product names get translated, legal terms drift from one document to the next, and the same correction gets made again and again.
Your terms, enforced
Required terms are checked after translation, including inflected forms like invoices or Rechnungen. Forbidden variants and do-not-translate names are caught and repaired.
Every segment shows its work
See which glossary entries, past translations and style rules shaped each sentence, with the model and prompt version on record.
Review only what's risky
Clean segments are marked clean. Reviewers start with the flagged ones, and every approved edit is reused in the next job.
How it works
From your glossary to a reviewed file.
- Step 1
Bring your assets
Import glossary spreadsheets and past translations. Syngram flags duplicates, conflicting targets and missing context before anything is applied.
- Step 2
Translate documents
Upload Word files, web pages, Markdown, JSON or XLIFF, in one or several target languages. Formatting, tags and placeholders come back intact.
- Step 3
Check and repair
Every segment is verified against your terms, numbers and tags. Misses are repaired; anything still uncertain is flagged, never passed silently.
- Step 4
Review and reuse
Reviewers work side by side in a bilingual editor. Approved segments go straight into your translation memory.
Provenance
Every segment shows its work.
When legal or compliance asks why a sentence reads the way it does, the answer is already on record: the terms, past translations and rules behind it, the checks it passed, and who approved it.
The invoice must be paid within 30 days.
Die Rechnung ist innerhalb von 30 Tagen zu bezahlen.
- Glossary
invoice → Rechnungfrom Legal terms, version 4- Memory
- Phrasing reused from an approved segment in Master services agreement 2025
- Style
- Legal profile: formal register, German number format
- Checks
- Required terms 1 of 1 · numbers intact · tags intact · no forbidden variants
- Approved
- By the language lead, with model and prompt version in the audit log
Languages
Seven languages, each one tested.
Right-to-left text is fully supported. A language is labelled tested only once it has its own evaluation set, so more languages are added as they pass.
- EnglishThe invoice must be paid within 30 days.
- EspañolLa factura debe pagarse en un plazo de 30 días.
- FrançaisLa facture doit être réglée sous 30 jours.
- DeutschDie Rechnung ist innerhalb von 30 Tagen zu bezahlen.
- 简体中文发票须在 30 天内付清。
- 日本語請求書は30日以内にお支払いください。
- العربيةيجب سداد الفاتورة خلال 30 يومًا.
File types
Works with the files you already have.
At launch
- Word (.docx)
- HTML and web copy
- Markdown and plain text
- JSON strings, including i18next and ICU
- XLIFF 1.2 and 2.0
Next
- PowerPoint, including speaker notes
- Excel, text cells only
Later
- Digital PDFs
- Scanned documents and images
Your data
Your content stays yours.
No training on your content
Content is routed only to model providers that don't train on customer data.
A full audit trail
Who translated, edited and approved each segment, with which model and asset versions.
Paperwork ready
A data processing agreement and a public list of subprocessors come with launch.
Questions
Questions language teams ask.
What is Syngram?
Syngram is an AI translation platform for companies with an in-house language team. It turns your glossaries, translation memory and style guides into a translation engine that follows them, checks every segment against them, and sends only risky content to human review.
How is it different from DeepL or Google Translate?
General-purpose engines translate well and can take a glossary as input. Syngram also checks every segment after translation against your terms, numbers, tags and style rules, repairs what it missed, flags what it can't fix, and records which assets shaped each sentence. Approved edits are reused automatically in later jobs.
Does it replace our translators?
No. Syngram writes the first draft and does the checking. Your language team owns the glossary and style rules, reviews the flagged segments and approves the result.
Which languages and file types does it support?
At launch: English, Spanish, French, German, Simplified Chinese, Japanese and Arabic, and Word, HTML, Markdown, JSON and XLIFF files. PowerPoint and Excel follow, then PDFs and scanned documents.
Can we bring our existing glossary and translations?
Yes. Import glossaries from spreadsheets (CSV or XLSX) and translation memory from TMX. You can also upload a past document together with its translation, and Syngram aligns the two into memory.
Is our content used to train AI models?
No. Syngram routes content only to model providers that don't train on customer data.
When can we start?
Early access opens to a small number of language teams first. Leave your email and we'll write to you once, when it opens.
Early access
Be one of the first language teams on Syngram.
Early access opens to a small number of teams first. We'll write to you once, when it opens.