Private AI chat

A language model running entirely on your own graphics card. No account, no server, no upload — and once the model is downloaded, it works with the network off.

What it is

This is Qwen3.5-4B, an Apache-2.0 model, compiled for WebGPU and run through WebLLM. The weights are served from this site and cached in your browser; nothing you type is sent anywhere. You can watch the network tab and see for yourself — which is a stronger guarantee than any privacy policy.

It is a 4-billion-parameter model, so it is not ChatGPT. It reasons about text you give it perfectly well and it does not reliably know facts — it will invent a date or an attribution with total confidence. Treat it as something to think with, not something to look things up in.