Amazon Bedrock introduced a Managed Knowledge Base that replaces the traditional customer‑managed setup where teams had to provision the vector store, tune embedding models, and configure indexing. The change means the service now handles ingestion, storage, indexing, and retrieval, while still allowing teams to control content and application logic.
What Changed: From Customer‑Managed to Managed Knowledge Base
In the previous model, engineers were responsible for scaling the vector store, selecting embedding models, and planning capacity for each data source. The new Managed Knowledge Base abstracts those responsibilities, providing automatic scaling, hybrid search (keyword + semantic) that is always enabled, and a managed embedding model with an optional managed reranker. Teams can still supply their own Bedrock embedding model, but the managed reranker requires the managed embedding model.
Why Engineers Should Care
The managed offering reduces the operational burden of maintaining a vector store and index settings, which translates to lower maintenance effort and potentially lower cost. Built‑in capabilities such as native connectors (S3, SharePoint, Confluence), ACL awareness, agentic retrieval, and integration with AgentCore Gateway add functionality that would otherwise require custom development. Higher default quotas (200 data sources per knowledge base, 10,000 knowledge bases per account per region) also simplify large‑scale deployments.
Architectural and Operational Implications
- Infrastructure simplification: The vector store, indexing pipeline, and scaling logic are managed by Bedrock, allowing teams to remove those components from their IaC templates.
- Connector usage: Native connectors ingest directly from supported sources. After an initial full sync, incremental syncs capture only changes, reducing data transfer and processing time.
- Access control handling: ACL awareness imports document permissions and filters retrieval results based on the signed‑in user. Since ACL awareness does not perform authentication, applications must still verify user identity and pass it to the Retrieve call.
- Agentic retrieval: Multi‑step queries can be handled iteratively, and a single request can span multiple knowledge bases, which may affect request latency and error handling strategies.
- Cross‑account access: Resource‑based policies let other AWS accounts call the knowledge base, scoped to specific IAM roles and actions such as
bedrock:Retrieve. This enables shared data assets without exposing the underlying store. - Migration workflow: The Agent Toolkit’s
amazon-bedrockskill provides a migration guide that performs read‑only discovery, generates a step‑by‑step plan, and can create the managed knowledge base alongside the existing one or output an IaC template for pipeline integration.
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
Adopt the Managed Knowledge Base when you want to offload vector store management and gain immediate access to built‑in connectors, ACL filtering, and agentic retrieval. Use the Agent Toolkit migration guide to map existing configuration, generate an IaC artifact, and validate retrieval before cutting over traffic. Ensure your application continues to authenticate users and passes identity information to Bedrock, as ACL awareness does not replace authentication. Evaluate cross‑account policies if multiple teams or accounts need shared access, and monitor quota limits when scaling the number of data sources or knowledge bases.