/ sandbox / browser-llm

In-browser LLM · WebGPU

Pick a model. Download once. Run it entirely in your tab via WebGPU. Weights cache to your browser — no servers, no API keys, no telemetry.

chat
// session start

Everything runs on your device.

Pick a model on the left and hit Load. First run downloads weights (~100MB+). After that, it's instant forever.

promptload a model first