Instrument anything that can curl.

A Heartbeat monitor is just a URL your job hits every run. If the pings stop, you get alerted. No SDK, no agent to install, no account to make. Every example below is a real command you can paste today.

Create a monitor. Go to /new for a form, or self-serve from an agent:

curl -sS https://golemreach.com/heartbeat/api/monitors/self-serve \
  -H 'Content-Type: application/json' \
  -d '{"name":"nightly-etl","period":3600,"kind":"push"}'

The response gives you a ping_url like https://golemreach.com/heartbeat/ping/<key> and a public https://golemreach.com/heartbeat/status/<name> page. For a server you cannot make outbound calls from, set "kind":"pull" and a "target" URL — Heartbeat probes it for you.

Make your job hit that URL on every run. One line, appended to the job. Success and failure are both signal.

Watch the status page. It shows state, 30-day uptime %, schedule and last success. Embed the badge https://golemreach.com/heartbeat/badge/<name>.svg anywhere.

cron Append the ping to the job so a silent miss is impossible:

*/15 * * * * /opt/etl/run.sh \
  && curl -sS https://golemreach.com/heartbeat/ping/KEY >/dev/null

Missed runs (exit non-zero before the &&) never ping, so they page — exactly the failure you care about.

systemd timer In the .service, ping on success:

[Service]
ExecStartPre=-curl -sS https://golemreach.com/heartbeat/ping/KEY >/dev/null
ExecStart=/opt/worker/run.sh
ExecStartPost=-/bin/sh -c 'curl -sS https://golemreach.com/heartbeat/fail/KEY >/dev/null || true'

Use ExecStartPost to record a fail when the main process exits non-zero.

GitHub Actions (scheduled) Workflows silently die — 60-day auto-disable, queue delays, dropped runs. A ping survives all of it:

on:
  schedule:
    - cron: '*/15 * * * *'
jobs:
  ping:
    runs-on: ubuntu-latest
    steps:
      - run: curl -sS https://golemreach.com/heartbeat/ping/KEY >/dev/null

Store the URL as a repo secret and reference ${{ secrets.HB_URL }} to keep it out of logs.

Unattended opencode / Claude Code loop Loops crash overnight and grep-and-hope is not a monitor. Ping once per loop turn:

# in the loop wrapper, after each turn:
curl -sS https://golemreach.com/heartbeat/ping/KEY >/dev/null

Better: a Stop-event HTTP hook (opencode plugin / Claude Code settings.json hook) that pings on every turn boundary. If the loop stalls, pings stop, you get paged. No dashboard to babysit. Full opencode / Claude Code recipe ↗

Windows Task Scheduler The unserved half — bulk-arm every enabled task with one ping URL. Full recipe + LastTaskResult table:

powershell -Command "Invoke-RestMethod 'https://golemreach.com/heartbeat/ping/KEY'"

Monitor Windows Task Scheduler ↗

Run-gate (kill switch) The mirror of a heartbeat: check the gate BEFORE work. One flip halts every loop at its next boundary:

[ "$(curl -s -o /dev/null -w '%{http_code}' https://golemreach.com/heartbeat/gate/KEY)" = 200 ] || exit 0

We dogfood this: our own fleet stops via a hosted gate.

Get alerts where you already are

Webhooks — Slack / Discord / generic: set a webhook URL when you create the monitor. JSON POST on every down/up transition.
Nostr DMs (ext#8, unique in this category) — add a nostr npub to the monitor and you get an encrypted NIP-17 gift-wrapped DM when it dies, plus an optional public obituary note. No email relay needed.
Email — pending an SMTP relay; disclosed honestly on pricing.
Egress self-check (ext#14) — a selfcheck monitor accepts signed verdicts from a probe script on your box and pages on transitions only (trusted→challenged), so a provider silently starting to 403 your IP never goes unnoticed.

Honest posture

Heartbeat is a dead-man switch, not a metrics platform. It tells you one thing reliably: the pings stopped. It does not sample your job's internals. For that, pair it with your existing logs. Free while in beta; a $5 USDC pass lifts the rate limits — see pricing.