What changed
Amazon Bedrock Managed Knowledge Base now supports single-shot Retrieval Augmented Generation (RAG) and retrieval-only requests through AgenticRetrieveStream. These modes allow applications to match retrieval effort to each request without building separate systems for simple and complex questions.
Single-shot RAG searches knowledge bases once and uses a foundation model to generate an answer, while retrieval-only requests return matching passages without generating an answer.
What this means in practice
Applications can now use the same API for both straightforward lookups and complex queries.
For example, a banking operations employee can use single-shot RAG to quickly retrieve a checklist for opening a business account and agentic retrieval to compare requirements across different business structures.
The new modes provide better support for multi-part questions, follow-up searches when information is missing, and answers with supporting evidence. Developers can configure retrieval-only requests when an employee needs to review source passages directly.
Limitations and availability
The expanded maxAgentIteration range now allows values from 0 to 5. maxAgentIteration=0 returns passages without generating an answer, while maxAgentIteration=1 is for single-shot RAG. Values from 2 to 5 support agentic mode, which includes query decomposition, evidence evaluation, and iterative searches.
The service manages the planning, retrieval, evaluation, and response-generation workflow through one API, reducing the need for custom orchestration. The changes are available for applications using Amazon Bedrock Managed Knowledge Base.
