Agent Governance
Register, monitor, and control autonomous AI agents.
Every agent starts at registration, enters PENDING_APPROVAL, and must be approved before activation. Agents follow a strict lifecycle with 6 states and 8 transitions.
Lifecycle States
| State | API key active |
|---|---|
PENDING_APPROVAL | No |
ACTIVE | Yes |
PAUSED | No |
SUSPENDED | No |
REVOKED (terminal) | Invalid |
REJECTED (terminal) | Invalid |
Suspend agent
You suspend an agent on its detail page. Suspending refuses every request that names that agent; the dashboard then shows the status the server returns, and the agent stays suspended after a reload. Suspending does not change the agent's API key. To stop every request that uses the project key, use the project emergency stop.
HIGH-Risk Requirements
HIGH-risk agents (EU AI Act) require: a completed FRIA before activation, a transparency disclosure, and human oversight.