Amazon Bedrock adds agentic retrieval to managed knowledge bases
Amazon Bedrock now offers agentic retrieval for its managed knowledge bases, letting developers break complex queries into sub‑queries, evaluate evidence, and re‑search if needed. The AWS Machine Learning Blog walks through using LangChain and the langchain‑aws package to call either the standard Retrieve API or the AgenticRetrieveStream API.
Key points
- Amazon Bedrock Managed Knowledge Base now supports agentic retrieval via AgenticRetrieveStream API
- LangChain’s langchain‑aws package exposes both standard and agentic retrieval functions
- Agentic retrieval improves recall on MuSiQue, especially for hardest multi‑hop questions
The tutorial covers prerequisites such as an AWS account with Bedrock access, IAM roles for the knowledge‑base service and API caller, and an S3 bucket for documents. It explains how to create a knowledge base, ingest data, and query it with both retrieval paths, including cost considerations. Agentic retrieval improves recall on the MuSiQue multi‑hop benchmark, especially for hard questions, and supports guardrails in BLOCK mode.
Practical guidance is provided: use standard retrieval for single‑intent questions to keep latency low and costs down; use agentic retrieval for multi‑part queries to get more accurate answers. The post also shows how to read trace events, handle scoring differences, and tune the maxAgentIteration parameter. The feature is available in us‑east‑1 and other Bedrock‑managed‑knowledge‑base regions.
The story so far
4 episodes →- Amazon Bedrock adds agentic retrieval to managed knowledge basesthis story
Agentic retrieval with LangChain and Amazon Bedrock Knowledge Bases
AWS Machine Learning Blog · 5 October 2026
Loading the full article…
This text was published by AWS Machine Learning Blog and written by Manideep Reddy Gillela. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Anthropic tests internal plugin system for Claude Code · 1 src
- Amazon Bedrock AgentCore Evaluations tests multi-agent systems for explainability and accuracy · 1 src
- AWS, Google Cloud launch AI agent tools; FTC investigates OpenAI, Anthropic · 2 src
- Enterprises need runtime controls to manage autonomous AI agents · 1 src
- Cohere launches North 2 with redesigned agent harness and cross‑session memory · 1 src
Comments
via GitHub Discussions