Monitoring That Keeps Working Without AI

When a service stops answering, I want the monitoring system to tell me what it saw and when it saw it. That’s the part I need to be able to count on, even if the AI service is having an outage of its own.

That’s how I think about Panoptes, the monitoring system I’m building. It collects events and runs read-only checks. The core rules and alerts can keep running on their own. AI can help summarize what’s happened or triage an event that the rules leave uncertain.

Keep the evidence close

If a service misses several checks in a row, I want to be able to see those checks and the rule that raised the alert. Then I can decide what to investigate. A summary is useful when a lot happens at once, especially if I can jump from it to the events it’s describing.

Panoptes shows whether a rule, AI, or a fallback classified an event. If AI can’t produce a summary, the dashboard can still put together a basic one from the data it already has. I want the person looking at the screen to understand how it reached that result.

A snapshot of the Panoptes public status view showing individual service states.
A snapshot of the public service status page powered by Panoptes.

You can see some of that work on the 404ops public status page, which gets its service status from Panoptes.