I Built AgentPulse Because I Was Tired of 3AM Pages
Last month my VPS ran out of disk space at 2:47 AM. Same log rotation failure. Same /var/log filling up. Same fix I’ve done a dozen times before.
That’s when I decided: if the fix is always the same, why am I the one doing it?
The Breaking Point
I had monitoring. Good monitoring. Uptime checks, process monitoring, disk alerts — the whole stack. I knew within seconds when something was wrong. The problem wasn’t detection. The problem was that I was the remediation layer.
3AM pages for disk space. 4AM pages for a crashed nginx. 6AM pages for an OOM process. Each time, I’d SSH in, run the same three commands, and go back to bed.
After the twelfth time, I started asking: why isn’t this automated?
The Market Gap
I looked at auto-remediation tools. They exist:
- Resolve.ai — Used by Coinbase and DataStax. Pricing? “Contact sales.”
- Dynatrace — $70+/host/month with complex licensing.
- PagerDuty Runbooks — Two products, two price tags, enterprise-oriented.
What about the solo dev running 3 VPS boxes? The indie SaaS founder with 5 servers? The small team that can’t justify $500/month in monitoring?
They’re stuck between:
- Free tools (Uptime Kuma, Netdata) — great alerts, no auto-fix
- Enterprise tools (Datadog, Dynatrace) — auto-fix exists, but the bill makes you cry
There was nothing in the middle. So I built it.
Building AgentPulse
The first version was simple: a Python agent that monitored my servers and ran shell commands when things went wrong. It worked, but it was messy. No baseline learning, no approval gates, no dashboard.
So I rebuilt it properly:
- A verify-or-escalate decision loop — every fix is simulated first, checked against hard safety rules, executed, then re-measured. If it didn’t hold, the agent escalates to me instead of retrying blindly.
- Approval gates — auto-fix, ask-first, or alert-only per action type, alert-only by default
- Baseline learning — the agent learns what’s statistically normal for each server and flags deviations
- Webhook alerts — get notified in Slack, Discord, or anything with an HTTP endpoint, but only when you actually need to act
- Hard safety rules in code — never sweep system paths, never touch a process without approval, allowlisted services only
(The read-only fleet console exists in source but is not deployed publicly. Native Telegram alerts and SSH brute-force detection also remain on the roadmap.)
The Install Experience I Wanted
I hate tools that take an afternoon to configure. AgentPulse is designed for a short setup, but public packaging and clean-host installation are still release gates:
Installation status: public self-serve packaging is not yet released. Do not run the draft installer on a production host. Review the current release gates.
Start in alert-only mode, review what the agent discovers, then promote safe remediation policies once the first server looks right.
The Pricing I Wanted
No “contact sales.” No per-GB billing that balloons unpredictably. No 14-page pricing calculator.
- Starter — $29/mo (1 server, alerts + approval-gated fixes)
- Pro founding — C$99/mo (up to 5 servers when fleet ships; auto-remediation, baseline learning, onboarding help)
- Business — $299/mo (finite server limit confirmed before purchase, custom policies, priority support)
The Result
I haven’t been paged for a fixable issue in weeks. AgentPulse handles disk cleanup and service restarts automatically, and queues runaway-process calls for a one-command approval. I still get alerts for things that actually need my attention — but those are rare.
If you’re running Linux/macOS hosts and tired of being the remediation layer, join the paid beta:
Installation status: public self-serve packaging is not yet released. Do not run the draft installer on a production host. Review the current release gates.