Governance covers how you control and audit AI agents across Agent Factory — confining what an agent's tool calls can do while it runs, and auditing what it actually did against your Postgres data afterward.
Access control
Access control confines what an agent's tool calls can do, and records every allow/deny decision, by mapping named purposes to Postgres roles enforced by Postgres's own privilege system. This is a built-in AIDB feature.
Register the agents and tools you want to govern before assigning them a purpose.
Register a purpose that resolves to a Postgres role, assign it to an agent, and review what happens when that agent runs — see Purposes and Enforcement.
Auditing
Audit what AI agents did against your Postgres data using EDB Agent Governance, a separate, standalone application, not a feature of AIDB: reconstruct agent sessions from the telemetry aidb records, including the governance decision behind each step. Postgres query logs from HM or Loki are supported as a legacy source.
Agents running outside AIDB can also feed this same audit trail directly, by sending their own OpenTelemetry traces, logs, and metrics to the OTLP receiver — see EDB endpoints.
Install EDB Agent Governance and connect it to a Postgres database running the aidb extension before auditing a session — see Installing Agent Governance and Connecting an aidb database.
Inspect the recorded steps and available execution metadata for a session — see Session detail.