AWS: Denodo + Amazon Bedrock AgentCore for customer 360

The post builds on a previous article in the series that used Stardog's knowledge graph as the semantic layer over Amazon Aurora and Amazon Redshift. This time the semantic layer is the Denodo Platform, and the source mix adds an on-premises PostgreSQL database, so the federation problem extends beyond an Amazon Virtual Private Cloud. The customer 360 schema stays the same, but the rewards_account table is moved from Aurora to the on-premises cluster, representing data from a third-party application. Denodo plans the federated query at execution time, pushing joins and aggregations to each source and combining the results inside Denodo, so no data moves between engines. The architecture also addresses identity: the end user's identity flows from the enterprise identity provider through AgentCore Identity On-Behalf-Of token exchange to Denodo, where Global Security Policies enforce row and column filters per caller. This design keeps access control at the semantic layer rather than in the agent's code or the model, which the post suggests reduces the risk of bypass through prompt injection. The post also notes that Denodo exposes the same view over multiple protocols, including JDBC, ODBC, GraphQL, and OData, and that Denodo joined Apache Ossie (incubating) in March 2026, a vendor-neutral specification for semantic metadata, though it is at an early stage.