Build a semantic layer for agentic AI on AWS with Strategy Mosaic and Amazon Bedrock AgentCore

This article demonstrates how to build a semantic layer for agentic AI using Strategy Mosaic and Amazon Bedrock AgentCore, enabling AI agents to query across Amazon Redshift and Aurora with governed access and business logic.

• Model a Customer 360 domain across Amazon Redshift and Aurora using two linked Mosaic models with no ETL required
• Define metrics using base, advanced, compound, and time-intelligence expressions with named filter objects and level qualifiers
• Expose the semantic layer to Strands agents through Mosaic MCP server with end-to-end user identity propagation for row and column-level security
• Choose between live pushdown to sources or in-memory execution for agent traffic based on performance and cost requirements
• Publish governed definitions to Power BI, Tableau, notebooks, and custom apps; export to Apache Ossie format for interoperability
• Enforce access control through OAuth 2.0 identity propagation from enterprise IdP through AgentCore to Mosaic engine
• Use Bedrock prompt caching to reduce token costs for repetitive agent queries against stable semantic models

Strategy Mosaic provides a dimensional semantic layer that serves multiple protocols (MCP, SQL, XMLA, REST) from one definition, enabling reliable agentic analytics with strong reasoning, trusted business context, and secure execution across live data sources.