Six clients. One me. No on-call rotation. If you freelance or run an agency and servers came "as part of the deal," this is the routine that keeps my phone silent.
The year I didn't have it, I lost a retainer. Client's site died at 1:40 AM on a Saturday. I found out from an angry DM at 9 AM. That is twelve hours of downtime with me as the last line of defense — because nobody was watching while I slept.
The routine, in the order that actually matters
1. Every client gets one canary, not a dashboard. A 30-line script that hits their login page and their checkout (or their top-3 URLs) every 5 minutes from OUTSIDE their network. Dashboards lie; a text message doesn't. If you can't get a text out of your monitoring, you don't have monitoring.
2. The dead-man's switch on the canary itself. The scariest failure mode isn't noise — it's the monitor that silently stops. Mine sends a heartbeat to a free healthchecks.io endpoint every run; if the heartbeat stops, THEY text me. A watchdog that watches the watchdog. This one habit is worth more than every paid APM I've trialed.
3. Weekly 5-minute self-audit, automated. Every Monday my script emails each client a one-pager: what was checked, what broke, what got fixed, disk headroom, backup age. It takes zero minutes of my week and it's the single biggest reason they renewed. Clients don't churn over outages they saw handled; they churn over silence.
4. Restore drills beat backups. A backup you've never restored is a theory. Quarterly (automated calendar reminder), I restore one client's backup to a scratch box and time it. The one time I found a corrupt backup this way paid for the entire routine.
5. Disk space alarms at 80%, not 95%. 95% is when you start the fire drill. 80% gives you a calm Tuesday fix. Full disks are still the #1 dumbest outage I see in small-client infra.
None of this needs a pager tool, an APM budget, or a second hire. It needs the scripts wired once and then watched by something that never sleeps.
I packaged my exact 24/7 solo-ops layer — canaries, dead-man's switch, the auto-generated client evidence reports, and the restore drill — as a self-serve starter pack:
https://hive80lab.gumroad.com/l/automation-starter-pack
And the full always-on operations system (the whole "watch the watchdog" architecture):
https://hive80lab.gumroad.com/l/agent-ops-24-7
Do step 1 today — it's 30 lines and it will catch the next Saturday outage before your client does.
Top comments (0)