Bedrock's managed knowledge base goes GA

Amazon Bedrock Managed Knowledge Base is now generally available, a fully managed retrieval service meant to replace the usual do-it-yourself stack of connectors, parsers, vector stores, and access-control logic that teams assemble to ground agents in enterprise data. Setup through the AWS console requires no model selection to start; AWS says teams get a first retrieval running in minutes, versus the days or weeks typically needed to hand-build a comparable pipeline.

The service ships with six built-in connectors covering the usual enterprise document stores, S3, SharePoint, Confluence, Google Drive, and OneDrive, plus a web crawler, and a direct ingestion API for anything outside those sources. Later syncs process only new or changed documents rather than reprocessing everything. Managed Knowledge Base also layers real-time access-control-list checks on top of pre-retrieval ACL filtering, so a query only surfaces documents the requesting user or agent is actually permitted to see. Teams that want more control can still configure their own embedding models, rerankers, and chunking strategies rather than accepting the defaults.

What it means for operators

The pitch is squarely at teams currently paying the integration tax of stitching together a RAG pipeline themselves: separate infrastructure, separate billing, separate rate limits, held together by internal glue code. It lands the same week Smartsheet published its own case study on building a remote MCP server on AWS and Grok 4.3 arrived on Bedrock, more evidence AWS is racing to make agent infrastructure a managed-service line item rather than something every team assembles itself. If document-level permissions are a hard requirement for your use case, the built-in real-time ACL enforcement is the detail worth testing first, since that’s usually the part hand-rolled pipelines get wrong or skip entirely.