Amazon Bedrock Managed Knowledge Base introduces Agentic Retrieve, adding user memory on top of document retrieval. The feature extracts durable facts from past conversations, selects those relevant to the current question, and injects them into the retrieval request with per-user and per-document isolation. It targets support assistants, in-product document chat, and internal policy assistants.
Agentic Retrieval
智能体检索
Retrieval paradigm where the model orchestrates multi-step retrieval and memory injection instead of single-shot vector recall.
Curated Intelligence
Amazon Bedrock Managed Knowledge Base now supports tabular search, letting agentic retrieval look up values and run calculations across rows in CSV, XLSX, and XLS files, then combine results with facts from related documents to answer multi-step analytical questions. The capability reuses the existing AgenticRetrieveStream request and works across mixed sources such as SharePoint and S3.
Amazon Bedrock Managed Knowledge Base now supports retrieval instructions that let teams define per-tenant answer voice, format, and retrieval policy as a versioned managed resource. Previously, enforcing a fixed answer style required disabling response generation and maintaining prompts in every application.
Amazon Bedrock Managed Knowledge Base now supports single-shot RAG and retrieval-only requests alongside its existing agentic retrieval. Developers can match retrieval effort to each query, switching between fast single-pass search and deeper multi-step search without building separate systems.
This guide demonstrates building enterprise agentic retrieval on Managed Amazon Bedrock Knowledge Bases using AWS CDK. It addresses traditional RAG limitations by introducing a reasoning agent and provides deployable code samples.
Amazon Bedrock Knowledge Bases introduces 'agentic retrieval,' using a planner to decompose complex multi-dimensional questions into sub-queries to address intent loss in traditional single-vector RAG. Integrated with LangChain, this feature aims to improve coverage and accuracy for compound queries while comparing costs against standard retrieval.