What changed
Amazon Bedrock has launched a managed knowledge base (Type: MANAGED) that handles ingestion, storage, indexing, and retrieval automatically. This replaces the previous need for customers to manually set up and manage vector stores like Amazon OpenSearch Serverless. The AgenticRetrieveStream API now performs multi-turn planning, retrieval, and returns cited answers as a first-class service feature.
What this means in practice
The managed knowledge base allows for agentic retrieval, where a reasoning agent decides which knowledge base to consult, retrieves information, and composes a grounded answer with citations. This is available only on the managed knowledge base, which includes auto-scaling and managed embedding and reranking with service-managed models. The solution now includes two Amazon CloudWatch dashboards and two forms of evaluation (on-demand and continuous) as part of the deployment.
Limitations and availability
Agentic retrieval is only available on the Managed Knowledge Base. The customer-managed (DIY) option remains available for those who want to use their own vector store, such as Amazon OpenSearch Serverless, for specific compliance or data-residency needs. The observability and evaluation features are built into the design, making them integral to the system rather than add-ons.
