Razin builds AI that works on your phone, in your browser and on your own machines, in Arabic and English. Your data stays where it already is.
Each product does one thing properly. Together they cover AI on the device, AI in the browser, and AI that reads documents.
OrbPrivate, offline AI for Android. Chat, ask about photos and documents, and talk by voice, with no account and no cloud.
DeputyAn agent in your browser that turns any page into a typed API for other agents. One call instead of screenshots and clicks.
Deputy against Playwright MCP, filling the same seven-field booking form with the same prompt. Both succeeded.
| Measure | Deputy | Playwright MCP | Result |
|---|---|---|---|
| Claude tokens | 133,366 | 216,888 | 1.63× fewer |
| Total cost | $0.1861 | $0.2730 | 1.47× cheaper |
| Wall clock | 17.2 s | 26.1 s | 1.52× faster |
Same seven-field form, same prompt, both through claude -p. Source: Deputy measurements
LinkAn Android app for the coding agents on your own computer. Start a task, review the exact change and approve it, or open a terminal, from your phone.
General-purpose OCR for images and scanned PDFs, on your own hardware. We're making it especially good at Arabic, dots and diacritics included.
{ "text": "Quarterly report 2026", "reading_order": 0 }| Configuration | Strict CER | Median s/image |
|---|---|---|
| Tesseract fast, ara+eng | 52.80% | 0.199 |
| Tesseract best, ara+eng | 42.03% | 0.288 |
| PaddleOCR PP-OCRv5 + Arabic recognizer | 15.23% | 1.888 |
Arabic, 14 synthetic pages, CPU, 6 September 2026. Diagnostics, not accuracy on your documents.
We help teams run models privately, read Arabic documents, connect agents to the web they already use, and run coding agents on their own machines.
Choose, quantize and run models on phones, laptops and your own servers, with GPU and NPU acceleration and offline-first apps.
Built on what we learned shipping Orb
OCR and extraction for documents in Arabic, English and mixed scripts, evaluated on your own pages before anything goes to production.
Built on our work on Razin OCR
MCP servers, WebMCP, and turning the web apps you already run into tools agents can call safely. Coding agents such as Claude Code and Codex on your own computers, directed from a phone over SSH.
Built on what we learned building Deputy and Link
Reproducible benchmarks on your data, so model, vendor and hardware choices rest on numbers.
Built on our evaluation harness
Models run on the phone, in the browser or on your servers. Your data stays put unless you decide otherwise.
Every number we publish comes with the method that produced it. If we haven't measured it, we don't claim it.
Arabic isn't a translation layer. We test on Arabic text, mixed scripts and right-to-left layouts from the start.
Limits get listed as clearly as features. A developer preview is labelled as one.