A dead man's switch for your AI agents

Why agents need a different kind of monitoring

Error alerts assume something runs and throws. Scheduled agents mostly fail in two quieter ways:

1. They never run. A platform incident, an expired token, a paused schedule, a laptop that was asleep.
2. They run and produce nothing useful. An empty result set, a refusal, a "sorry, I couldn't access that", a summary of zero items.

People running fleets of agents describe opening each one every morning just to check it ran. A dead man's switch flips that: each agent checks in when it finishes, and you only hear about the ones that didn't.

How it works in Did It Run?

• Expect a run: "every weekday by 9:15am", a cron expression, or "every 15 minutes", with a timezone and grace window.
• Check in: ping a URL (with /start and /fail variants) from anything that can make an HTTP request. Email check-ins are coming soon.
• Check the outcome: optional minimum item count, minimum size, or required keyword turn empty runs into alerts.
• Get told: missed, late, failed and empty alerts by email, Slack, Discord or Telegram — and a recovery notice when it's fixed.
• Read the roll-call: one daily digest across every agent and platform.

Works where your agents live

Recipes for ChatGPT scheduled tasks, Claude Code, n8n, Zapier, Make, GitHub Actions and cron.