AWS launches Bedrock Managed Knowledge Base for enterprise RAG

AWS used its New York Summit to launch Bedrock Managed Knowledge Base, a fully managed offering designed to remove the heavy lifting from building enterprise retrieval-augmented generation (RAG) pipelines. The service bundles native data connectors, 'Smart Parsing' to prepare multi-format documents, and an 'Agentic Retriever' capable of handling multi-step queries—abstracting away the storage, retrieval, embeddings, and foundation-model plumbing developers normally wire together by hand.

The pitch is zero-custom-code RAG: developers point the service at proprietary data and get grounded, citable answers without managing vector databases or embedding pipelines. Crucially, it offers flexibility to choose different embedding and re-ranking models, and it integrates with AgentCore Gateway so the knowledge base can feed AWS's broader agent stack. The Smart Parsing component targets the messy reality of enterprise data—PDFs, tables, mixed formats—that typically degrades RAG quality.

Reaction in developer communities was sparse but positive, with praise for native MCP integration and the ability to skip custom RAG infrastructure entirely. Some practitioners pushed back, noting they still prefer the flexibility of running their own vector stores on Postgres or Redis rather than ceding control to a managed black box. That tension—convenience versus control—will define adoption.

The launch is one piece of a much broader agentic push at Summit NYC, alongside AgentCore web search, Continuum security, and continuous modernization tooling. Together they signal AWS's strategy: rather than competing head-on at the frontier-model layer, it is racing to own the enterprise plumbing that makes agents production-ready. Competitively it squares off against Microsoft's Azure AI and Google Cloud's Vertex offerings. Watch for pricing details, real-world accuracy benchmarks, and whether enterprises trust a managed retriever with their most sensitive proprietary data.