AWS launches Managed Knowledge Base for Bedrock with 6 enterprise connectors and automatic ACL enforcement

Amazon Web Services launched Managed Knowledge Base for Amazon Bedrock in general availability, a fully managed retrieval solution that eliminates the need to provision and scale vector databases manually. The service includes six native data connectors and handles document parsing, storage, and access control automatically.

Native connectors and real-time access control

Managed Knowledge Base ships with connectors for Amazon S3, Microsoft SharePoint, Atlassian Confluence, Google Drive, Microsoft OneDrive, and a web crawler. The service includes a direct ingestion API for documents outside these sources. On subsequent syncs, the system processes only modified or new documents.

The service performs real-time ACL checks by verifying permissions directly with the authoritative source at query time, rather than relying on cached or mapped ACL data. According to AWS, pre-filtered documents exist only for the duration of the API call and are not visible to LLMs or users.

Syngenta Group is using the service to enable employees to create knowledge bases that sync data from SharePoint and Confluence. MRH Trowe deployed it for an internal AI copilot spanning thousands of documents in English and German.

Multimodal parsing and automatic storage

The service handles visual content documents (PDFs, PPT/PPTX, DOCX) up to 500 MB, audio files up to 2 GB, and video files up to 10 GB. It automatically selects parsing strategies for tables, charts, diagrams, mixed layouts, and media without configuration.

Managed Knowledge Base provides unified storage that auto-provisions capacity and scales from gigabytes to terabytes. Hybrid search combining keyword and semantic retrieval runs continuously with no separate index configuration. Data is encrypted at rest and in transit using AWS KMS keys.

Setup and retrieval APIs

Setup requires three API calls: create the knowledge base, add a data source, and start ingestion. AWS provides defaults for embedding models and chunking strategies, though developers can override these with fixed-size chunking or no chunking for pre-processed documents.

The service offers two retrieval APIs: Retrieve returns ranked source chunks with relevance scores for direct lookups and FAQ-style questions. A second API (not fully detailed in the source material) handles more complex queries requiring multi-hop reasoning.

OpenAI confirmed it is using Bedrock Managed Knowledge Base's RAG capabilities to ground inference for millions of users with customer context.

What this means

This release consolidates the enterprise RAG pipeline into a single managed service, eliminating weeks of infrastructure setup for teams building AI agents. The real-time ACL verification at query time addresses a significant security gap in knowledge retrieval systems that typically cache access permissions. The 500 MB document size limit for PDFs and 10 GB for video positions this as infrastructure for internal enterprise search rather than public-facing applications with larger content libraries.