I bulit an AI firewall because I couldn’t see what my own machine was sending out

Actually It’s simple. I just had no clue what Claude Code, Claude Desktop, or other AI assistants and providers were doing under the hood.
So I built a tool that monitors all AI traffic leaving my MacBook. I can block and allow providers, and see in real time which tools get called and stuff like this.
It runs completely on the device, so no requests are routed through the cloud.
It lets me block the AI search on Google or the integrated Gemini in Gmail, for example. But also, because of the monitoring, I’ve had a much easier time debugging AI-related implementations I’ve built on the side.
Looking forward to hearing your opinions on that, and happy to answer questions about the technical stuff. You can also try it without any login or anything at patronus.studio/download.