← zyth.io

A chat assistant running on your GPU

A Zyth program runs a language model in this tab, on your own graphics card through WebGPU. No server answers you: once the model is downloaded, it even works offline, and nothing you type leaves your device.

 
first word –
on
ready in

Pick a model

It is downloaded once and kept by your browser for next time.

Getting the model ready

Starting…
Small models make mistakes. Qwen2.5 by Alibaba Cloud (Apache-2.0), run by Zyth.