I built WebBrain because most AI sidebar extensions send every page you look at to someone else’s servers.
How it handles privacy:
- Point it at llama.cpp, Ollama, LM Studio, vLLM etc. on your own machine and no page content leaves it
- No telemetry, no tracking, no account needed
- Read-only “Ask” mode is the default; the agent has to be switched to “Act” to click or type, and it asks before consequential actions
- GPL-3.0, so all of this is auditable
It’s a Chrome/Edge/Brave/Firefox extension. Cloud models (OpenAI, Claude, OpenRouter…) are optional, bring-your-own-key.
Happy to answer questions, and I’d appreciate critique of the threat model — especially around giving an agent access to a logged-in browser.
5 comments
log in to comment.
You should be aware that the Feddiverse is overwhelmingly anti-ai and you will probably be downvoted quite severely.
The only real question I have about this is that it seems kinda pointless, chatting with a single webpage, making an AI fill out forms.
Why not just read the page yourself and fillout the form yourself? Most browsers have had forms autofil functionallity for many many years now. What advantage does the AI bring to the browsing experience?
No, you didn’t.
Thank you for sharing this. I’ll give it a try at some point. Unfortunately Lemmy is still mostly on the denial phase of the LLM revolution or on a path to become a full on Ludite extremist heaven. But I’m glad there’s still people working towards securing our privacy in an inevitably LLM dominated future tech world.