Chat is an Arabic and English assistant that answers from the documents your staff upload.
Policies and procedures sit in PDF and Word files. Staff search them by hand.
Staff paste internal text into public chat tools to get an answer.
Staff work in Arabic and English. The assistant must work in both.
Upload PDF, Word and text files to a knowledge base.
Finds the five closest passages in your database. Sends them with the question to your model.
Streams back to the user. Chat saves it in the conversation history.
Chat uses the model endpoint that you configure. Anar supplies no models.
Answers stream word by word. Chat sends the last 10 messages as context.
Search, rename and delete conversations. Export them as JSON or Markdown.
See usage statistics for conversations in the console.
Upload PDF, Word and text files. A PDF needs a text layer.
Chat stores the vectors in your database. It finds the five closest passages.
Chat cleans Arabic text and splits it into sentences. It stores the cleaned form, not the original spelling.
Chat reads only the text layer today. Arabic OCR for scans is on the roadmap.
Chat tells the model to cite its sources. The user does not see the sources yet.
Most queries filter by tenant. One chat path does not check who owns the knowledge base.
Roles are admin, manager and user. The sign-up route needs a lock before a pilot.
| Capability | Status | Note |
|---|---|---|
| Chat, history, export, analytics | Shipped | On in every install. |
| Knowledge bases, upload, retrieval | Shipped | PDF text layer, Word and text files. |
| Arabic text handling | Partial | Stores the cleaned text, not the original. |
| Source citations | Partial | Not shown to the user yet. |
| Tenant isolation, sign-up | Partial | Gaps to close before a pilot. |
| Arabic OCR, labels, erasure | Roadmap | Planned. Not built. |
- Streaming chat and history
- Knowledge bases and retrieval
- Export and analytics
- Tenant check on every knowledge base
- Arabic OCR for scanned files
- Arabic search that keeps the original text
- A label on each AI answer
- Classification labels and clearance
- Erasure across vectors, exports and logs
- A published Arabic retrieval test
Chat can send every model call and every embedding call through the gateway. Gateway headers can also sign users in.
Chat can send its events to the Guard audit trail. Guard does not filter Chat answers yet.
A one-month pilot: we deploy in your environment and run it with one team.
anarlabs.ai/contact