An AWS machine learning blog post describes a reference architecture that combines Denodo Platform with Amazon Bedrock AgentCore to provide governed access to distributed enterprise data for an agentic AI application. The walkthrough uses a customer 360 example spanning Amazon Aurora, Amazon Redshift and an on-premises PostgreSQL database.

According to the post, Denodo acts as the semantic layer and exposes a unified customer_360 view through VQL, its SQL dialect. The example joins customer profile data from Aurora, order history from Redshift and rewards-program information from the on-premises database using a shared customer identifier. It also defines a reusable lifetime customer value metric.

The architecture uses the Denodo MCP Server to expose governed views to an AgentCore-hosted Strands agent. A foundation model on Amazon Bedrock plans queries, while AgentCore Gateway handles inbound authentication and routes tool calls. The post says AgentCore Identity exchanges the end user’s token for a scoped token representing that user to Denodo. Denodo Virtual DataPort then resolves the caller’s role, applies policies and plans the federated query.

The post says Global Security Policies can enforce row filters and column masking at query time. It also states that source data remains in the underlying systems rather than being moved between engines; VDP sends rewritten native queries to the sources and combines the results. The described Denodo deployment options are Agora, a fully managed cloud service, or self-managed installations on Amazon EC2 or Amazon EKS.