AI agent monitoring for runs that stop without telling you

Your scheduled agent or LLM job pings a URL when it finishes. If the ping is late, or the job reports a failure, UptimeIO alerts you.

How scheduled agents fail quietly

An agent that runs on a schedule, such as a nightly research summary, a support-ticket triage pass, or a batch of LLM calls over new data, has no page to load and no port to check. When it stops working, it usually stops in one of three ways.

It never starts, because the scheduler, the server, or the workflow tool did not fire it. It starts and hangs, waiting on a model call or a tool that never answers. Or it exits with an error after a rate limit or an expired API key. Logs only help if someone reads them, and an error handler only helps if the process gets far enough to run it.

A heartbeat monitor catches all three. The job tells UptimeIO when it finishes. If that message does not arrive on time, you hear about it.

How it works

  1. Create a heartbeat monitor

    Choose the Heartbeat type, name it after the agent, and set how often it runs plus a grace period for slow runs.

  2. Mark the start of a run (optional)

    Call the /start URL when the agent begins. UptimeIO then records how long each run took. A start call never opens an incident on its own.

  3. Ping when the run ends

    On success, ping the heartbeat URL with exit_code=0. On failure, call the /fail URL with the exit code, and an incident opens right away.

  4. Get alerted when a run is missing

    If no ping arrives within the interval plus the grace period, the run never started or never finished, and your alert channels are notified.

What you configure

The same settings as any heartbeat monitor. Wrap the agent command once and every run reports in.

  • Expected interval: how often the agent runs, from 1 minute to 24 hours. The shortest interval is 5 minutes on Free and 1 minute on Pro and Scale.
  • Grace period: extra time after the interval, from 1 second to 24 hours. Agent runs vary in length, so allow for your slowest normal run.
  • Exit code and duration: pass exit_code and duration_ms on the ping, or call /start and let UptimeIO work out the duration.
  • Optional metadata: a POST ping can carry up to 10 small JSON values, such as items processed or tokens used.
BASE=https://heartbeat.uptimeio.com/heartbeat/YOUR_HEARTBEAT_TOKEN
curl -fsS "$BASE/start" > /dev/null
./run-agent.sh
code=$?
if [ "$code" -eq 0 ]; then curl -fsS "$BASE?exit_code=0" > /dev/null
else curl -fsS "$BASE/fail?exit_code=$code" > /dev/null; fi
Example: wrap a scheduled agent so success pings and failure reports

What alerts you get

A missed run opens an incident after the interval plus the grace period. A /fail call opens one immediately, titled "Job Reported Failure", so you can tell a crash from a run that never happened. Each event in Recent Heartbeats shows its exit code and duration. The incident resolves after 3 consecutive on-time pings.

  • Email, Slack, Telegram, Pushover, and webhooks, on every plan (Pushover on Free needs your own app token)
  • Webhook requests can be signed with HMAC-SHA256, so a webhook can start your own follow-up
  • There is no SMS or phone-call alerting

Which plans include it

Heartbeat monitoring, including /start, /fail, exit codes, and durations, is available on every plan. Free includes 50 monitors with a 5-minute shortest interval. Pro includes 100 monitors and Scale includes 300, both with a 1-minute shortest heartbeat interval.

Heartbeat URLs are rate limited; see the docs for the limits. Treat the URL as a secret, and keep it out of prompts and logs your agent can read back or share.

Frequently asked questions

What is AI agent monitoring?

For a scheduled agent, it means checking that each run actually happens and finishes. The agent reports to a heartbeat URL at the end of every run, and you are alerted when a report is missing or says the run failed.

Can UptimeIO tell if my agent produced a bad answer?

No. UptimeIO does not read or judge the output. If your job can detect a bad result, such as an empty summary or zero items processed, have it call the /fail URL, and UptimeIO alerts on that.

What happens if the agent hangs?

A hung run never sends its final ping, so an incident opens once the interval plus the grace period has passed. Set the grace period a little longer than your slowest normal run.

Does this work with n8n, cron, or GitHub Actions?

Yes. Anything that can send an HTTP request can ping the heartbeat URL: a shell script, a cron entry, an n8n HTTP Request node, or a step in a scheduled workflow.

Is it available on the Free plan?

Yes. Heartbeat monitoring, including /start and /fail, is on every plan. On Free the shortest expected interval is 5 minutes.

Related

Start monitoring in a few minutes

Start monitoring free

Free plan · 50 monitors · No credit card required