AWS launches Bedrock AgentCore with managed knowledge base, web search, and payment infrastructure for AI agents

AWS Expands AgentCore with Managed Knowledge Infrastructure

Amazon Web Services has released new capabilities for Bedrock AgentCore designed to connect AI agents to enterprise knowledge bases, live web data, and paid information sources. The update addresses three infrastructure gaps: accessing scattered organizational data, retrieving current information beyond training cutoffs, and enabling payment for premium content.

Managed Knowledge Base Handles Enterprise Data

Bedrock Managed Knowledge Base, now available on AgentCore, connects to data sources including SharePoint, Google Drive, Confluence, S3, and internal wikis. AWS manages the vector store, embedding models, re-ranking, and rate limiting.

The system uses what Amazon calls an "agentic retriever" that plans queries across knowledge bases, connects concepts across documents, evaluates intermediate results, and re-ranks before returning answers. According to AWS, this approach provides broader coverage for complex, multi-part queries compared to traditional retrieval-augmented generation (RAG) systems.

Web Search Using Amazon's Infrastructure

AgentCore's new Web Search tool provides access to public web information while keeping queries within AWS's security boundary. The feature runs on the same search infrastructure used by Alexa, Amazon Quick Suite, and Kiro.

The search system combines web results with Amazon's proprietary knowledge graph, which includes structured entity data, verified facts, and real-time information such as stock prices and sports scores. Queries remain within the customer's AWS environment with no external vendor required.

Payment System for Premium Data Access

AWS introduced two complementary capabilities for handling paid content: AgentCore payments (announced in preview in December, pricing not yet disclosed) and WAF AI traffic monetization (now generally available, pricing not yet disclosed).

AgentCore payments enables agents to discover paid services and content, access them, and complete payment within their execution loop. WAF AI traffic monetization allows content providers to control agent access and receive compensation. Because both run on AWS infrastructure, providers using WAF automatically recognize verified AgentCore agents.

Enterprise Adoption

Sony Group Corporation is building an enterprise AI agent platform on AgentCore where teams across business units can develop and share agents. According to Masahiro Oba, Senior General Manager at Sony, the company's enterprise knowledge is distributed across SharePoint, Confluence, and S3, including complex documents with charts and tables.

"Now that Bedrock Managed Knowledge Base and Web Search are available in AgentCore, we can equip agents with advanced retrieval and live web grounding with a consistent governance model, without building these capabilities from scratch," Oba stated.

Optimization and Monitoring

AWS also announced new optimization capabilities that convert production traces into improvement signals, though full details were not provided in the announcement. The company emphasized that agent failures often produce no error signals, surfacing only through customer complaints after affecting thousands of sessions.

What This Means

AWS is building comprehensive infrastructure for enterprise AI agents rather than just offering model access. The managed knowledge base eliminates months of pipeline engineering work that previously blocked enterprise deployment. Web search integration addresses the training data cutoff problem that limits agent usefulness for time-sensitive queries. The payment infrastructure attempts to solve the economic problem of agents accessing premium data sources—critical for financial services, research, and market intelligence applications where the best information isn't free. However, AWS provided no specific pricing details, benchmark comparisons, or performance metrics for any of these features, making cost-benefit analysis difficult for enterprises evaluating adoption.