◐

Browser LLM Chat

A chat assistant running 100% in your browser — no server, no API key, fully private. Powered by WebGPU.

Checking WebGPU support…

Choose a model

The model is downloaded once (a few hundred MB to ~2 GB) from a free CDN, cached in your browser, then runs on your own GPU. The first load is the slow part; after that it is instant.

Starting…

WebGPU is not available in this browser 😕

This demo needs WebGPU to run the model on your graphics card. Try:

The latest Chrome or Edge on desktop (best support).
On Linux you may need to launch with --enable-unsafe-webgpu.
Check your status at webgpureport.org.

Model loaded and ready. Ask anything 👇

System prompt

Temperature: 0.7

Changing the model? Reload the page and pick a different one before loading.

Built with WebLLM · runs on-device via WebGPU · nothing is sent to any server.