Translate sends Arabic and English text to the model you choose, with your approved terms and your past translations, then routes the result to a reviewer.
A team translates the same sentence again because nobody can find the last version.
Two translators use two words for one legal term.
Drafts move by email. Nobody can see which version was approved.
Arabic or English. Pick a domain: legal, administrative, technical, medical, financial or general.
Finds past translations and approved terms, and builds the instruction for the model.
The model you configure writes the translation. Translate stores it for review.
Anar supplies no models. Text stays in your network when you point Translate at a model in your network.
Similarity
Above the 0.80 threshold. Translate sends the stored Arabic to the model as a starting point. An exact match is reused without a model call.
Output of Translate's matching code on these sentences, run 18 November 2026. The memory entry is an example.
Glossary entry
الرخصة التجارية
Keep one glossary for each team. Add terms in bulk and see how often each one is used.
The instruction Translate built for this text, run 18 November 2026. Lines are shortened where marked.
Your model writes the first translation of each segment.
A translator edits the segment. The editor locks it while they work.
A reviewer approves the segment. Every revision stays on record.
Source and target side by side, one segment at a time.
See who changed a segment and when.
The model rates its own translation. Use it as a hint.
Set a gateway address and every translation call goes through it, with your keys, budgets and logs.
Translate sends an event for each translation. You can check a text for personal data before you send it.
One memory and one set of glossaries for every team, behind the same controls as the rest of the platform.
| Capability | Status | Note |
|---|---|---|
| Text translation, memory, glossaries, TMX | Shipped | Arabic and English, six domains. |
| Segment editor and review | Shipped | Lock, revisions, draft to review to approved. |
| Documents and batches | Partial | .txt and .docx in, plain text out. Up to 50 texts a batch. |
| Quality score | Partial | A self-rating by the model, not a validated metric. |
| Personal-data check | Partial | Runs when you ask. It does not run on each translation. |
| Term check, protected numbers, more languages | Roadmap | Planned in this order. |
- Arabic and English, six domains
- Memory with TMX import and export
- Glossaries for each team
- Segment review and approval
- Check that the output used each term
- Keep numbers, dates and names unchanged
- Keep the document format
- Post-editing to ISO 18587
- Languages beyond Arabic and English
A one-month pilot: we deploy Translate in your environment and load your memory and glossaries for one team.
anarlabs.ai/contact