Agent Fleet
A small fleet of always-on AI agents with self-healing operations, shared memory, and human escalation.
A handful of always-on agents running across my machines and a VPS, wired
together so they can actually be left alone.
The interesting parts
- Shared memory — every agent reads and writes the same
MuninnDB store, so what one learns the others know. - Self-healing — health checks feed a heal-escalation chain; failures that
can't be fixed automatically page me through a beacon service and Telegram. - Private mesh — nodes talk over Tailscale; nothing agent-facing is exposed
to the public internet. - Gateway pattern — a single dockerized gateway fronts agent communication,
with config, restarts, and deploys handled over SSH and compose.
Why
Partly because it's useful — the fleet handles memory upkeep, monitoring, and
publishing chores. Mostly because running agents as infrastructure — with
uptime, alerting, and operational discipline — teaches you things that toy
demos never will.