No model loaded
This is a one-time download. The model is cached in your browser afterward, so future loads are instant. Everything runs on-device — nothing is uploaded.
Depending on the model and your connection, this can take a while — don't close the tab.
This is a demo of client-side (in-browser) AI chat. Every model — text chat, image captioning, and image Q&A — runs entirely on your device using WebGPU or WebAssembly. There is no backend server, no API key, and no account.
Model weights download once from Hugging Face or MLC's CDN and are cached in your browser (the Cache API) for instant reuse afterward. Chat history is saved only in this browser's local storage. Nothing you type, and no image you attach, is ever uploaded anywhere — inference happens on your own CPU/GPU, so it keeps working even offline once a model is loaded.
Source: https://github.com/janusdotai/localai
Provider support on this device