Every (personal) agent (like Muse, Instinct, Grokbot, Poke) asks you to connect your email, calendar, and tools again. Permissions are intransparent and every provider ends up building the same integrations.
We think there needs to be an independent tool that holds all connections to third party companies.
That way you can keep auth separate, audit what an agent does, set fine-grained permissions and easily switch between the agent apps.
I built this over the last few days with my friend Konsti. It's open source and we’ve put a lot of care into the self-hosting experience.
Now my Instinct has the same tools as Claude Code and Codex on my laptop.
What are your thoughts about this and how are you handling this today?
Comments URL: [https://news.ycombinator.com/item?id=50064514](https://news.ycombinator.com/item?id=50064514)
Points: 1
source & further reading
github.com — original article
A Modern Agentic AI Operating System
Show HN: Popcorn: a security workbench of disposable microVMs
Tidy – Open-Source Paper