Amazon Bedrock Managed Knowledge Base Adds Tabular Search for Analytical Questions over Tables
AWS has introduced tabular search for Amazon Bedrock Managed Knowledge Base, enabling agentic retrieval to natively analyze and answer questions over CSV, XLSX, and XLS files. This expands the platform’s retrieval capabilities from just text to structured table data, supporting calculations and multi-source analytical queries.
What changed?
Amazon Bedrock Managed Knowledge Base now supports tabular search, letting agentic retrieval answer questions based on data in CSV, XLSX, and XLS files. When a question requires calculations or lookups across a table, the system runs model-written code in a secure sandbox over the entire table, retrieves relevant values, and combines them with facts from related documents. This is available through the AgenticRetrieveStream interface and requires no additional configuration. The standard Retrieve API remains unchanged and does not perform table computations.

Why does it matter to an everyday developer?
Developers can now build applications that answer complex, analytical questions that rely on structured data within table files, not just text documents. This is particularly relevant for domains where knowledge is embedded across spreadsheets and accompanying policies or documentation, such as finance or operations. The answer includes both calculated results and citations to relevant files, improving traceability and trust. No changes are needed in ingestion pipelines: supported files from Amazon S3, Microsoft SharePoint, Confluence, and other connectors are handled automatically.
There are important limitations. Responses that rely on tabular search are slower, because the answer requires one or more additional retrieval steps and code execution. Tabular search is not a replacement for traditional business intelligence or reporting tools. Schema definition is not supported; the system operates directly on raw uploaded files. Correct retrieval still depends on identifying the right table, so specifying a file name in the query may improve results.
What can the developer do now?
- Start asking analytical or calculation-based questions over ingested CSV, XLSX, and XLS files via agentic retrieval with AgenticRetrieveStream.

