The AWS Machine Learning Blog outlines an agentic retrieval approach for Amazon Bedrock Managed Knowledge Base. The post addresses the limitations of classic retrieval when handling multi-part questions and introduces the AgenticRetrieveStream API. It describes how the API works, including request construction and trace parsing, and provides guidance on when to select agentic retrieval over the standard Retrieve API.
Why it matters
Multi-part questions can be difficult for standard retrieval methods, and the post frames agentic retrieval as an alternative intended for these cases within the Bedrock Managed Knowledge Base.
Who should care
Developers and teams building retrieval workflows on Amazon Bedrock who need to decide between the AgenticRetrieveStream API and the standard Retrieve API.