# Amazon Bedrock Managed Knowledge Base adds single-shot RAG and retrieval-only modes

published: 2026-11-09T21:05:39.177Z
topic: search
source: https://aws.amazon.com/blogs/machine-learning/single-shot-and-agentic-retrieval-with-amazon-bedrock-managed-knowledge-base/
url: https://aiagentnews.io/stories/amazon-bedrock-managed-knowledge-base-adds-single-shot-rag-and-retrieval-only

Amazon Bedrock Managed Knowledge Base now runs the AgenticRetrieveStream API in three modes, chosen per request through the maxAgentIteration parameter, whose range has been extended to 0 through 5. A value of 1 runs one retrieval and one generated answer, while a value of 0 returns passages without calling a model.

AWS says the Amazon Bedrock Managed Knowledge Base now supports single-shot retrieval augmented generation and retrieval-only requests through the AgenticRetrieveStream API. Single-shot RAG searches a knowledge base once and has a foundation model write an answer from the passages it finds. Retrieval-only requests return matching passages with no generated answer and make no foundation model calls.

Agentic retrieval remains available for values from 2 to 5. In that mode, a foundation model splits a question into sub-queries, runs several searches, evaluates the evidence and refines its searches when information is missing. Each iteration equals one foundation model call, and the final answer generation uses the last one, so a value of n allows up to n minus 1 planning rounds. The initial retrieval that runs before the first planning step does not count toward the limit.

Setting generateResponse to true with a maxAgentIteration of 0 returns a 4xx error, because generating an answer requires a model call. AWS's illustrative banking example uses a value of 1 for a checklist lookup and 4 for a comparison across policy documents. All three modes support multiple knowledge bases, and the application, not the service, chooses the mode for each request.

## Why it matters if you build agents

Agent builders can send simple lookups and multi-step questions through the same API and set the iteration limit per request, instead of maintaining a separate retrieval path. Single-shot mode uses one model call, which can reduce latency and model usage compared with agentic planning, and passage-only requests use no model calls at all.

AWS notes that a higher iteration limit does not always produce a better answer, so teams should test each limit against their own documents, including questions the documents cannot answer, before choosing a default.