AtScale and Amazon Bedrock AgentCore for a governed semantic layer
In brief
- AtScale builds a dimensional semantic layer on Amazon Redshift for a customer 360 model, exposed via MCP to a Strands agent on Amazon Bedrock AgentCore.
- Identity, roles, and row-level security are enforced at the semantic layer, not in the agent’s prompts.
- The semantic model is versioned in YAML on Git and publishes data simultaneously via SQL, XMLA, and MCP.
AWS and AtScale introduce a solution to integrate a governed semantic layer with AI agents via Amazon Bedrock AgentCore. According to the article, AtScale models customer data into a dimensional model on Amazon Redshift and exposes it to a Strands agent through the MCP protocol. Identity, roles, and row-level security are enforced at the semantic layer, not in the prompts.
AtScale’s semantic model relies on a declarative language (SML) versioned in Git. It enables publishing data simultaneously via SQL, XMLA, and MCP, ensuring consistent answers for the agent and existing BI tools. The architecture uses Amazon Bedrock with Anthropic’s Claude Sonnet 4.6 to plan queries, while Amazon Bedrock AgentCore manages authentication and orchestration.
This approach is part of a series focused on building semantic layers for agentic AI. It targets enterprises with scattered customer data, such as retailers with customer records, orders, and products spread across multiple systems. The solution avoids pre-computed joins and centralizes analytical data in Amazon Redshift, synchronized with operational systems via AWS Database Migration Service or zero-ETL.
What it changes for you
- Retail & E-commerceRetailers can consolidate scattered customer data (orders, products, rewards) into a governed semantic layer, reducing join errors and improving consistency in AI agent responses.
This summary gives you the essentials. The full article, its figures and quotes are on Artificial Intelligence.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error