Idea Lab · Oct 15 2026 · Fleet winner · Tool/build

FleetHeartbeat

Each agent sends a small "I'm alive" ping on a schedule. A monitor colors each node green, yellow or red by how long it has been silent, and restarts a node that stays red too long. This simulation runs a fleet clock at 10× speed so you can watch failures and restarts happen.

Simulated fleet

Simulation · no real servers are contacted

06:00:00

Ping every (s) Yellow after silent (s) Red after silent (s) Restart after red for (s) — proposal: 90 Max restarts per node per hour, then page a human

Event log

Claimed by the proposal, not verified

"Build time: 3 days." Plausible for the basic version; a restart that is safe (doesn't fight systemd or kill a half-written package) takes longer.
"Eliminates manual presence checks before the 9 AM brief." Only if the monitor itself is monitored; a dead monitor shows nothing, not red.
Port 8899 is listed for this WebSocket server, but the package already uses port 8899 for the Governance Hub. One of them needs a different port or a path on the same server.

What to try

Crash an agent: it goes yellow, then red, then gets restarted and returns to green.
Hang one: the process is "up" but stops pinging. A restart fixes it, which is the case where auto-restart earns its keep.
Break for good: restarts never help. Watch the restart cap stop the loop and page a human instead of restarting forever.
Flaky network: 60% of pings are dropped. With a 30 s yellow line it flickers; raise yellow to 45 s and the noise mostly goes away.
A real build: a cron or systemd timer on each VPS posting to one endpoint over Tailscale, a 50-line server, and the restart done by systemd on the node itself.