What changed
AWS has introduced a Managed Knowledge Base for Amazon Bedrock, which automates the ingestion, storage, indexing, and retrieval of data. This service handles scaling as content grows and includes hybrid search capabilities combining keyword and semantic matching. The Managed Knowledge Base also features a managed embedding model, a managed reranker, and a built-in parser for multimodal file types. Users can still bring their own embedding model when creating the knowledge base.
What this means in practice
The Managed Knowledge Base simplifies the migration process by providing a migration guide in the Agent Toolkit for AWS. This guide helps users plan the migration, create the new knowledge base, and update their application’s retrieval calls. The service includes native connectors for sources like Amazon S3, Microsoft SharePoint, and Atlassian Confluence, and supports real-time permission checks for content access. Additionally, it allows integration with AgentCore Gateway, enabling clients that support the Model Context Protocol (MCP) to query the knowledge base. Resource-based policies also allow applications in other AWS accounts to use a team’s knowledge base, with permissions scoped to specific roles.
Limitations and availability
The Managed Knowledge Base is available in specific AWS Regions, with some features not supported in AWS GovCloud (US-West). Higher default quotas are available, including up to 200 data sources per knowledge base and 10,000 knowledge bases per account, per Region. Pricing is based on storage and retrieval usage, with no minimum idle costs. The service does not support agentic retrieval, AgentCore Gateway integration, or resource-based policies on customer-managed knowledge bases.
