Local AI Chat runs a Gemma model directly on your machine through WebGPU, so every message and image you share stays on your device. Attach a screenshot or scanned page and it reads the text with in-browser OCR, then answers using the rest of your conversation as context. The model (about 1.9 GB) downloads once and is cached, so after that you can chat with no network connection at all. Nothing is sent to a server, logged, or used for training.
Yes. The model runs inside your browser tab on your own GPU. Your messages and any images you attach never touch a server, so there is nothing to leak, log, or train on.
Only for the first load, which downloads the model (around 1.9 GB) and caches it. After that the chat works fully offline.
Images are processed locally with OCR to pull out their text, which the AI then uses to answer you. The files stay on your device and are never sent anywhere.
Every tool runs entirely in your browser. Files are never uploaded — 100% private and free.