Check your AI agent register
Paste the rows from your register. In seconds you see which agents have no named owner, no stop switch, no human-approval rule, or no protected backups.
Private by design: the check runs in your browser. What you paste or upload is not sent to Lumis or anyone else, and nothing is saved. Close the tab and it is gone.
Needs the headers of our template. Get the template
What this check looks at
- Every agent has a named owner. A team name triggers a warning.
- An agent that can write, delete, send or pay has at least one action marked "human approval".
- Every agent has a stop switch: who can stop it, and how.
- Agents that change data name an action log and confirm backups they cannot reach.
- Systems and data, access level, status and a last-review date within three months (format YYYY-MM-DD).
This is a completeness check, not a security audit. It cannot tell whether what you wrote is true, and the owner and access checks work from wording, so they can be wrong. Treat it as a prompt for a conversation with the people who run each agent.
Columns it reads: Agent · Task · Owner · Systems and data · Access · Needs human approval · Stop switch · Action log · Backups out of reach · Status · Last review