SLM Chat WebGPU
Download a small language model to your browser and chat with it. The model, your prompts, and the generated text never leave your device.
Pick a model
Smaller downloads faster and runs on more devices. Larger answers better. All downloads are cached in your browser — pick once, chat anytime.
What happens when you load a model. Model weights are fetched from Hugging Face directly to your browser and cached in origin-private storage (OPFS). The WebLLM engine and WASM runtime are fetched from jsDelivr. From then on, inference runs on your GPU via WebGPU — no further network traffic.
Privacy & security.
Nothing you type is sent to any server. The model weights are static files
fetched from Hugging Face and cached on your device. No analytics on this
page, no third-party cookies, no account.
What this needs. A browser with WebGPU (Chrome / Edge / Brave on desktop today; Safari Technology Preview; Firefox with the WebGPU flag). ~0.4–3 GB of free storage per model.
What this does not do. Access your microphone, camera, location, clipboard history, or any file on your computer. The only permission requested is persistent storage, so the browser doesn't evict a model you downloaded.
What this needs. A browser with WebGPU (Chrome / Edge / Brave on desktop today; Safari Technology Preview; Firefox with the WebGPU flag). ~0.4–3 GB of free storage per model.
What this does not do. Access your microphone, camera, location, clipboard history, or any file on your computer. The only permission requested is persistent storage, so the browser doesn't evict a model you downloaded.