PostgreSQL

PostgreSQL handles the state categories that belong in a relational database rather than in OpenSearch or ClickHouse: entities with referential integrity requirements, configuration that must be consistent across multiple readers, and records where foreign key constraints enforce correctness. Agent configuration lives here because agents need to look up their own configuration reliably; identity anchors (the baseline identity vectors that define the starting point for identity drift measurement) live here because they must be pinned to specific agent versions; session state metadata lives here because sessions reference agents, policies, and identity records that must remain consistent; policy version history lives here because each policy snapshot must be attributable to a specific session and time. PostgreSQL is not used for the high-volume, high-throughput data paths - those go through Kafka and OpenSearch. It handles the relatively low-volume coordination and configuration data that requires the relational guarantees that columnar stores and search engines cannot provide.