AtScale and AWS detailed a Redshift-based customer analytics example where an agent queries versioned business definitions with identity-aware controls.
In a technical post dated October 20, 2026, AWS and AtScale described a customer analytics workflow that pairs AtScale’s semantic layer with an agent hosted on Amazon Bedrock AgentCore. The example uses a Strands agent and Anthropic Claude Sonnet 4.6 on Amazon Bedrock to answer questions against customer data modeled in Amazon Redshift.
AtScale’s model is authored in Semantic Modeling Language, an open YAML-based specification that can be versioned in Git. The same deployed model is exposed through SQL, XMLA and the Model Context Protocol (MCP), allowing an agent and business intelligence tools to use shared definitions. The example includes measures such as revenue, order count and units sold, plus calculated measures including customer lifetime value. Teams can revise a definition in the model and review the change through a Git-based workflow.
The walkthrough keeps analytical customer, purchase and product data in Redshift. Amazon Aurora PostgreSQL remains the operational system of record, with AWS Database Migration Service or Aurora zero-ETL to Redshift described as ways to keep the analytical copy current. AtScale plans dimensional queries as native warehouse SQL. Its aggregate manager can also route matching queries to a pre-built summary table rather than scanning the purchase fact table.
The agent is instructed to list available models, load the chosen model’s dimensions and measures, and then query using those definitions. The authors tested questions including top products by units sold and high-spending customers without rewards enrollment. They also compared results with an Excel PivotTable using the same model and reported matching totals for equivalent questions.
Access follows the caller’s identity. In the post’s example, an HR user’s model perspective includes Social Security number data, while a marketing user’s perspective does not expose that field. The reference deployment places AtScale on Amazon EKS and connects its MCP server to AgentCore Gateway. AWS presents this as an implementation guide in a series of semantic-layer examples, rather than as a new product launch.