Seer: open-source browser extension for local AI image descriptions via screen readers, using PaliGemma2 model with no cloud upload or API key required.

Seer is a browser extension that automatically describes images without alt text using PaliGemma2-3b-mix-224 (a vision-language model), processing all images locally on the user's computer with nothing sent to a server.

Setup requires installing dependencies (fastapi, uvicorn) and running a local daemon on http://127.0.0.1:11435, then loading the extension in Chrome via developer mode; the model files total ~2.4 GB and require ~3 GB RAM with no GPU needed.

Built for people using screen readers (NVDA, JAWS, VoiceOver) who encounter images on websites without descriptions; no internet, account, or subscription required after initial setup.

Licensed under Apache 2.0 by Recursia Lab.