Governance

AI agent governance and audit

Governance on Achilles is enforced outside the model. Permissions live in code, so an agent cannot take an action the definition did not allow, even when the prompt would have.

Every turn records the definition that ran, the memory that was read, and a typed event for each tool, handoff, and branch. When a goal is missed, the trace shows where, and the fix maps back to the part of the definition that caused it.