01
Create an agent
Name it, set every 30 minutes or a cron expression, and choose a grace period. You get a secret URL.
Healthchecks for AI agents
StaleAgent is a dead-man's switch for scheduled agents — Claude Code, Codex, or a script of your own. Each agent gets a ping URL. If the ping does not arrive, you get one alert. When it recovers, you get one more.
The open-source wrapper sends exit code, duration, byte counts, and a hash of stdout. It never sends the prompt, the output, environment variables, or API keys.
01 — Setup
01
Name it, set every 30 minutes or a cron expression, and choose a grace period. You get a secret URL.
02
curl the URL from cron, or wrap the command: staleagent run <token> -- claude -p "triage inbox"
03
Missed, failed, empty, too slow, too expensive, or stuck repeating itself. A recovery note when it is healthy again.
02 — Alerts
No start or success ping by the schedule plus grace.
A /fail ping, or an exit code that is not zero.
A successful run that reports zero bytes.
A run still open, or finished, past the limit you set.
One run, or the last 24 hours, over the limit you set.
The same stdout hash several times in a row, or pings far faster than the schedule.
03 — Privacy
The hash cannot be turned back into the output. Turn it off with --no-hash.
04 — Pricing
Billing is not switched on. New accounts are on Free. The alert rules above already run.
$0
$6/mo
$12/mo