The job that never ran

Last updated 17 November 2026

By Mumen Tayyem, founder of Jabee

Published 17 November 2026.

Every alert you have wired up has the same blind spot. The backup script sends you a message when it fails. The deploy pipeline posts to Slack when it breaks. The sync job emails you on an exception. All of them are the job talking — so when the job does not run at all, nothing talks, and the silence reads exactly like success.

A dead man's switch inverts it. The job says "I ran" each time it finishes, on a schedule you declare up front, and the thing being monitored is the silence. Miss the check-in and somebody is told. It is the only shape of monitoring that can see a cron line somebody commented out three weeks ago, a container that stopped being scheduled, or a worker that died before it reached its own error handler.

The one line that does it

0 3 * * * /usr/local/bin/backup && curl -fsS -X POST \
  -H "Authorization: Bearer $JABEE_API_KEY" \
  "https://api.jabee.ai/api/v1/heartbeat/nightly-backup?every=24h&grace=30m"

The && is the whole design: check in only if the job actually succeeded. A ; there would report health every night while the backup failed every night.

every is required the first time and sets the schedule — 15m, 2h, 90s, 1d, or a plain number of seconds. Later check-ins inherit it, so a second caller cannot silently re-time somebody else's check. grace is how late it may be before anybody is woken: a job that usually takes four minutes and sometimes eleven wants grace=15m, not a pager at minute five.

When the deadline passes with no check-in, the alert arrives on WhatsApp — on the handset the business verified, from Jabee's own approved sender. There is no WhatsApp number to register, no business verification to wait out, and no message template to get through review, because the number and the template are ours and already approved.

In Python, if you would rather not shell out:

import os, urllib.request

def check_in(name, every="24h", grace="30m"):
    url = f"https://api.jabee.ai/api/v1/heartbeat/{name}?every={every}&grace={grace}"
    request = urllib.request.Request(
        url, method="POST",
        headers={"Authorization": f"Bearer {os.environ['JABEE_API_KEY']}"},
    )
    with urllib.request.urlopen(request) as response:
        return response.status          # 201 the first time, 200 after

The full file, with its tests, is examples/python/heartbeat.py; the Node equivalent sits beside it. Both run with nothing installed.

What this costs, and what the alternatives cost

The honest comparison, from each vendor's own published pricing page, read on 17 November 2026:

Checks includedPublished priceWhatsApp delivery
Healthchecks.io20$0 Hobbyist, $5 Supporter, $20 Business (100 checks)Yes — SMS and WhatsApp credits are included on paid plans
Cronitor5$0 Hacker, then $2 per monitor per month plus $5 per userEmail and chat on every plan; SMS on Business. WhatsApp is not named
Better Stack10 heartbeats on the free tier$0 free tier, $20/month for 10 more heartbeatsEmail, SMS and phone calls are named. WhatsApp is not
Jabee20$29/month, which also covers 500 alerts and approvalsYes, and nothing to register

Read that table honestly: if all you want is a free dead man's switch with email delivery, Healthchecks.io has a free tier with the same 20 checks and has had one for years. We are not going to pretend otherwise, and at $29 against $0 the price is not the argument.

What the $29 buys is the other half of the product. The same key that takes a heartbeat also sends alerts from your own code, and asks a person for a decision on WhatsApp with Approve, Skip and Edit buttons whose answer comes back to your system over a signed webhook. If the only thing you need is cron monitoring, buy cron monitoring. If what you actually want is for your systems to reach a person on the app that person already has open — and sometimes to wait for their answer — then the heartbeat is one of four things one subscription does.

Why silence is harder than it looks

Three details decide whether a dead man's switch is useful or just another thing that cries wolf.

The grace period is not optional. A job whose deadline is exactly its interval alerts on every slow run. Set grace to the longest honest overrun and the alert means something.

Recovery has to be asked for. Add recover=true and the next successful check-in after a missed one tells you it is back. Leave it off and you will be told when things break and never when they mend, which for a nightly job is the right default more often than people expect.

Twenty checks is the ceiling. One business may register 20. That is more cron than most businesses have, and well short of what a loop could create by accident.

POST /api/v1/heartbeat/<name>?every=15m&grace=5m&recover=true
GET  /api/v1/heartbeats                 # what you have registered
DELETE /api/v1/heartbeats/<name>        # stop watching one

Names are lowercase letters, digits, hyphens and underscores, up to 60 characters. The first check-in creates the check and answers 201; later ones answer 200.

Setup, field by field, is on the heartbeat documentation page. If what you want is the opposite direction — your code telling a person something, rather than the absence of a message doing it — that is alerts, the same key and the same handset.

Questions

Can a push-only alert service do this? No, and that is structural rather than a missing feature. Anything that waits for your job to send it something cannot distinguish a job that never started from a quiet night. Only a check-in with a deadline can.

What happens if my job runs late every so often? Set grace to the longest overrun you consider normal. The alert fires only when the check-in misses both the interval and the grace period, so a job that sometimes takes twice as long does not page anybody.

Do I need my own WhatsApp number? No. Alerts go out from Jabee's approved sender to the handset your business verified, so there is nothing to register and no message template to get through Meta's review. To reach several people, you connect your own WhatsApp number instead.

How many checks can one account have? Twenty. Each one is a row read by a sweep every half minute, and each outage sends a real message, so the ceiling is deliberate rather than a plan limit to be upsold past.

Is there a free way to try it? The first three sends are free, so you can watch an alert arrive on your phone before paying for anything. After that it is $29 a month, which includes the checks, 500 alerts and approvals, and the approval buttons.

Does it work from GitHub Actions or a CI job? Yes — it is one authenticated request, so anything that can run curl can check in. A CI job that is supposed to run nightly and silently stops being scheduled is exactly the failure this sees and a status badge does not.