AI agents / Supervision
superPi
A local UI for supervising multiple coding-agent conversations: tool activity, events, state and notes in one place.
The problem
When working with several agents, I need to keep context, activity and decisions visible without losing track across conversations.
My contribution
I built a local web interface for following Pi conversations. A Node.js server streams events via SSE and keeps notes for each conversation.
Available evidence
The README explains how to run it on Linux and describes the access options. The npm test command runs the checks offline without contacting an AI provider.
Usage and limits
It is experimental and not a sandbox. Agents can use the operating-system user’s permissions: do not expose it to untrusted networks, and do not treat Tailscale as authentication.