Anar Docs turns PDFs, scans and Word files into text, document types and key fields, with the model endpoint you configure.
Scans and photos hold the text. Staff type it again before a system can use it.
Someone reads each file to decide what type of document it is.
An assistant cannot answer from a document that nobody has turned into text.
A PDF, a Word file or an image, up to 50 MB. Seven file formats.
Reads the text, sets the document type, extracts fields and checks them.
Text, fields and chunks return as JSON. Chunks can go to Anar Chat.
Anar supplies no models. Scanned-page reading and field extraction use the model endpoint you configure.
Digital PDF and Word files give exact text and tables. No model is needed.
Scans and images go to your vision model. A local OCR engine is the fallback. Accuracy is not yet measured.
A batch request processes many files in one call.
Ten document types. Docs asks your model first, then uses Arabic and English keywords.
Finds Hijri and Gregorian dates, GCC national IDs, phone numbers and amounts. No model is needed.
Your model reads the first 4,000 characters. Long documents lose content. A pattern fallback runs without a model.
Docs checks for required fields. Today the check reads English field names only.
Docs splits the text into chunks and sends them to a Chat knowledge base. One call can process and send.
Shipped. Turn it on per install.Tested against a mock of Anar Chat. Chat accepts the chunks from an admin or manager account.
What Chat receives
- Text chunks ready for retrieval
- The document type
- Dates, IDs and other metadata
| Capability | Status | Note |
|---|---|---|
| Upload, digital text and tables, batch | Shipped | Seven formats. 50 MB per file. |
| Document type, GCC metadata | Shipped | Ten types. Arabic and English keywords. |
| Scanned-page OCR | Partial | Works. Accuracy is not yet measured. |
| Field extraction, validation | Partial | First 4,000 characters. Validation reads English field names. |
| Search | Partial | Keyword match on text. No semantic search. |
| Handwriting, redaction, review queue | Roadmap | Not built yet. |
- Upload, OCR and extraction
- Type and GCC metadata
- Export to Anar Chat
- Measured Arabic OCR accuracy
- Each field linked to its place on the page
- Personal data redaction
- Human verification queue
- Handwritten Arabic, stamps and seals
- Arabic interface
Docs can send chunks to a Chat knowledge base. Turn it on per install.
A Flow step can send a document to Docs for processing.
Procure can send tender text to Docs to list requirements.
Type and field calls can go through the gateway. Scanned-page reads do not yet.
A one-month pilot: we deploy in your environment and run it with one team.
anarlabs.ai/contact