AI & Data — Local Chat
Dozens of models, running in your browser
Pick an open model and chat: the weights are streamed once, then everything runs on your GPU through WebGPU. No API keys, no server, no account — your conversations stay in this browser's local storage.
New chat
First load of a model downloads its weights from the public MLC / Hugging Face CDN — between roughly 200 MB and 5 GB depending on the model. They are cached by the browser afterwards, so the next visit starts in seconds. Nothing is uploaded anywhere: prompts, answers and history never leave this device, and clearing your browser data removes them for good.