Amazon Bedrock Knowledge Bases
Give foundation models and agents contextual information from your company’s proprietary data sources to deliver more relevant, accurate, and customized responses.
Grounding agents and applications on enterprise data
Amazon Bedrock Knowledge Bases grounds generative AI in enterprise data, enabling applications and agents to make better decisions based on organizational information. For teams that want to get started quickly without managing infrastructure, Bedrock Knowledge Bases offers a fully managed option.
Amazon Bedrock Managed Knowledge Base eliminates undifferentiated infrastructure work. Connect unstructured company data sources and let the service handle the rest. We manage the vector store, embeddings and re-ranking models used during retrieval, and scalability including rate limits, so teams can focus on building rather than operating pipelines.
For teams needing full pipeline control, customizable options let developers bring their own vector stores and databases.
Overview
Managed Knowledge Base
Bedrock Managed Knowledge Base now offers a fully managed RAG service that connects to enterprise data sources like SharePoint, Confluence, Google Drive, OneDrive, S3, and Web Crawler. It automatically handles ingestion, vector storage, and retrieval optimization. Smart Parsing prepares diverse data types for accurate retrieval. Optimized defaults for embeddings, re-ranking, and foundation models mean developers don't have to deal with this complexity and can get started quickly.
Smart Parsing
Smart Parsing automatically determines the optimal parsing strategy for each data type and connector without manual configuration. It uses connector-specific data models that preserve metadata, embedded images, and thread context, while multimodal processing handles text, images, audio, video, and multiple languages through intelligent extraction and captioning. We manage the default settings on chunking and parsing strategies to ensure high-quality content extraction from the start, and customers can customize these to fit their specific needs. Because Smart Parsing handles all of this automatically, teams eliminate weeks of experimentation typically required to achieve production-quality retrieval accuracy.
Agentic Retrieval
Agentic Retriever is an optional feature purpose-built for complex, multi-step queries that require reasoning across enterprise data. It offers both simplicity and customizability. It abstracts away query complexity on the customer's behalf — automatically understanding intent, identifying which data sources to retrieve, and iterating to obtain the depth and breadth of information needed. This delivers accurate answers through a single API call with no custom orchestration code required. For customers using an existing search stack, they can bring their own models to tailor retrieval to their specific requirements.
AgentCore Integration
Bedrock Managed Knowledge Base integrates natively with Amazon Bedrock AgentCore as a pre-built target type, eliminating the need for Lambda functions and custom integration code. Through this integration, Knowledge Bases gain built-in observability with retrieval traces and performance metrics, session memory, and MCP compatibility — enabling customers to build end-to-end agentic systems. Bedrock Managed Knowledge Base can be configured as a retriever for agent frameworks including Strands Agents, LangChain, CrewAI, and LlamaIndex.
Price Predictability
Agentic Retriever delivers predictable per-query pricing that shields customers from the cost variability inherent in token-based approaches. Complex queries that require multiple reasoning iterations can lead to highly variable and escalating token costs when using general-purpose foundation models. Agentic Retriever minimizes this variability through a flat per-query pricing model powered by our proprietary retrieval model, significantly reducing overall cost compared to bring-your-own-model approaches — where per-query costs can be orders of magnitude higher with wide variance query to query.
Use cases
Ground support agents in product documentation, policies, and case history to deliver accurate, contextual responses that reduce resolution times across thousands of customers.
Give employees instant access to institutional knowledge spread across SharePoint, Confluence, Google Drive, and more, surfacing relevant answers from internal policies, documentation, and processes.
Reps investigate sales info using accumulated customer data and deal/opportunity records to accelerate pipeline and close deals faster.
Research manuals and past incident reports for operational problems, surfacing near-miss information to resolve issues quickly and prevent recurrence.
FAQs
General
Open allIt is is a fully managed Retrieval-Augmented Generation (RAG) service that grounds generative AI applications and agents in your company's proprietary data. It connects to enterprise data sources—including Amazon S3, SharePoint, Confluence, Google Drive, OneDrive, and web crawlers—and automatically handles ingestion, vector storage, embedding, re-ranking, and retrieval optimization. This enables foundation models to deliver more relevant, accurate, and customized responses without requiring you to manage infrastructure or build custom pipelines.
- Amazon S3 — cloud object storage
- Microsoft SharePoint — intranet and document management
- Confluence — team wikis and documentation
- Google Drive — cloud file storage
- Microsoft OneDrive — personal and shared files
- Web Crawler — public or internal web pages
Bedrock Managed Knowledge Base connects to a broad range of enterprise data sources, including:
The service uses Smart Parsing to automatically determine the optimal parsing strategy for each data type, preserving metadata, embedded images, and thread context while processing multimodal content such as text, images, audio, video, and multiple languages—all without manual configuration.
Agentic Retrieval is a capability within Bedrock Managed Knowledge Base purpose-built for complex, multi-step queries that require reasoning across enterprise data. Unlike traditional single-pass retrieval, Agentic Retrieval automatically understands user intent, identifies the most relevant data sources, and iterates through multiple retrieval steps to deliver comprehensive, accurate answers—all through a single API call with no custom orchestration code required. Customers can also bring their own models if they have an existing search stack.
Bedrock Knowledge Base natively integrates with Amazon Bedrock AgentCore as a pre-built target type—requiring no Lambda functions or custom integration code. It also serves as a data plane provider for popular open-source frameworks including LangChain, CrewAI, LlamaIndex, and Strands Agents. Built-in capabilities include observability with retrieval traces and performance metrics, session memory, and MCP (Model Context Protocol) compatibility, enabling developers to build production-ready AI agents faster.
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages