Experimenting with AI agents at home

This is my practical workbench for exploring what personal AI agents can do. The setup has involved a Raspberry Pi and an always-on Mac mini, with Hermes as an existing agent framework that I configure and use.

Telegram gives me a way to talk to the agents, Tailscale lets me reach the machines remotely, and an Obsidian vault gives the setup somewhere to keep useful context.

I’m interested in the everyday work around running agents: keeping them reachable, connecting the right context and finding tasks where they actually help. I’m also exploring running models locally. Running an agent at home and running all of its models offline are different things; the current setup also uses hosted models.

I’ll use this project to share the setup and what I learn as it changes.