In-browser playground
A real local LLM running on your device via WebGPU. The first load downloads the model (300 MB to 2.3 GB); it is cached afterwards. Prompts never leave your browser.
Models download once and are cached by your browser.
15 free messages left today.
Load a model, then ask something. Everything runs locally on your GPU: nothing is sent to a server. Try “Explain the KV cache like I'm 12.”