Chat
Ask anything. It runs on your GPU.
Your browser downloads an open model once, then your graphics card writes every answer inside this tab. What you type isn't sent anywhere, and chats are saved only in this browser. Pick a model below to start.
Every answer comes with a receipt: where it ran, how fast, and whether this tab made any network request while the model wrote. That check reads the browser's resource timing for this page and the model's worker, so it shows that nothing went out, not what a request would have carried.
Choose a model for this device
It runs on your graphics card. Sizes are the graphics memory each one needs. Nothing downloads until you press Load model.
The first load downloads the model from Hugging Face. After that it is cached in this browser and loads without downloading again. Only the model comes over the network; your messages never do.
Enter sends. Shift and Enter adds a line.