Get answers from tables with tabular search in Amazon Bedrock Managed Knowledge Base
Amazon Bedrock Managed Knowledge Base now adds tabular search, enabling agents to retrieve and compute over CSV, XLSX, and XLS files in a single request. The feature runs sandboxed code over the full table, then merges calculated results with facts from related documents to answer multi‑step analytical queries. Engineers can invoke this via the existing AgenticRetrieveStream API, allowing seamless integration of table‑driven logic into RAG pipelines.
⚡ Key Takeaways
- Tabular search supports CSV, XLSX, and XLS file formats.
- Retrieval is performed with the AgenticRetrieveStream request, which now also executes sandboxed code on the entire table.
- The approach solves long‑table and wide‑table retrieval gaps by fetching only one passage (the file) and processing all rows/columns in code.
- The trace output identifies which workbook the code read, providing provenance for the answer.
- The sandboxed execution isolates code, ensuring that only the intended table is accessed during inference.
Engineers can now build RAG agents that answer complex, calculation‑heavy questions without pre‑aggregating tables, reducing engineering effort and improving answer completeness for finance, operations, and other data‑rich domains.
✅ Practical Steps
- Use the AgenticRetrieveStream API to submit a query that references a CSV, XLSX, or XLS file stored in Amazon S3 or SharePoint.
- Inspect the returned trace to see which workbook was read and how the sandboxed code computed the answer.
Want the full story? Read the original article.
Read on AWS ML Blog ↗