MIRTEL

Did your agents do their chores?

Mirtel keeps one ledger for every scheduled agent you run: the Claude routine at 07:00, the ChatGPT task at 08:00, the n8n flow at 11:30. She knows what each one was supposed to do and when. She writes down what actually happened. And she notices the run that never started, which is the one nobody else notices.

She watches. She never touches. Mirtel never runs, retries or changes your agents, and she holds no passwords for the platforms they run on.

How it works, in four lines

  1. You tell Mirtel what an agent should do, when, and what a good run looks like. That's its contract.
  2. Mirtel gives you a link for that agent. You hand the link to the agent.
  3. At the end of every run, the agent reads the link and reports back: done, empty, failed, or needs you.
  4. Mirtel checks each expected run against what came in. Anything that didn't happen, or didn't go well, lands on your list.

Each agent gets one link of its own. It looks like this:

https://YOUR-MIRTEL-HOST/r/mrt_agent_…

If the agent opens it, it gets plain instructions written for it: its contract, its schedule, what counts as success, and exactly how to report. If it sends a report to the same link, Mirtel records it. So the only thing you have to tell an agent is:

Before you finish every run, read the link and do what it says.

Some apps can't send to a link but can use a connector. For those, the same key works as a connector address: https://YOUR-MIRTEL-HOST/mcp/mrt_agent_…. The setup pages say which apps that helps.

Mirtel shows you the link once, when you add the agent or rotate its key. Treat it like a password: anyone who has it can report as that agent. If it leaks, rotate it and the old one stops working.

Where to go next