Amazon Bedrock Managed Knowledge Base is now generally available
Build production RAG agents over enterprise data—no vector DBs or pipelines to manage, with multimodal support and agentic multi-hop retrieval built in.
View original announcement →Visual Summary
graph TD
A{{Bedrock Managed Knowledge Base}}:::announced
B((Enterprise Data Sources)):::external
C([Smart Parsing]):::feature
D([Managed Vector Storage]):::storage
E([Hybrid Search & Ranking]):::feature
F([Agentic Retrieval]):::feature
G(Amazon Bedrock AgentCore):::compute
H((AI Agents & Apps)):::external
I([ACL Enforcement]):::feature
B ==>|"S3, SharePoint, Confluence, etc."| A
A -->|"auto-selects strategy"| C
C -->|"embeds & indexes"| D
H ==>|"queries"| G
G ==>|"invokes"| A
A -->|"retrieves"| E
E -->|"re-ranks results"| F
F -->|"returns grounded answers"| G
A -.->|"filters at retrieval"| I
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
Amazon Bedrock Managed Knowledge Base is now generally available as a fully managed RAG service that eliminates the need to operate vector databases, data pipelines, or retrieval infrastructure. It connects to six native enterprise data sources, handles ingestion and storage automatically, and provides advanced retrieval capabilities including hybrid search, document ranking, and agentic multi-hop reasoning. The service integrates natively with Amazon Bedrock AgentCore, enabling teams to build production-ready AI agents grounded in enterprise data with built-in observability and auto-generated permissions.
How It Works
- Data Ingestion via Native Connectors: Six pre-built connectors (Amazon S3, SharePoint, Confluence, Google Drive, OneDrive, and Web Crawler) automatically sync data into the knowledge base without custom pipeline code.
- Smart Parsing: Automatically selects the optimal parsing strategy per document type—PDFs, PPTX, DOCX, scanned documents, and multimodal content (text, images, audio, video)—preserving metadata, embedded visuals, and thread context without manual configuration.
- Managed Vector Storage: AWS manages embedding models, vector indexing, storage auto-scaling, and rate limits, optimized for price-performance so developers never provision or tune a vector database.
- Hybrid Search & Document Ranking: Retrieval combines dense vector search with sparse keyword search (hybrid search) and applies re-ranking models to surface the most relevant results.
- Agentic Retrieval: An optional feature that decomposes complex multi-hop queries into sub-queries, iteratively retrieves across multiple knowledge bases, evaluates response sufficiency, and re-ranks results—all via a single API call with no custom orchestration code.
- Access Control Lists (ACLs): Document-level permission filtering is enforced at retrieval time for all connectors except Web Crawler, ensuring users only see data they are authorized to access.
- AgentCore & MCP Integration: Native integration with AgentCore Gateway allows any MCP-compatible agent framework to discover and invoke the knowledge base as a tool, with auto-generated IAM permissions and built-in observability including retrieval traces and per-knowledge-base metrics.
Why It's Important
- Eliminates Undifferentiated Heavy Lifting: Teams no longer need to build or maintain vector databases, embedding pipelines, chunking strategies, or re-ranking infrastructure, reducing time-to-production from weeks to hours.
- Enterprise-Grade Multimodal Support: The ability to ingest and retrieve across text, images, audio, and video in a single managed service addresses a growing need for richer, context-aware AI applications.
- Production Readiness Out of the Box: Built-in observability, retrieval traces, agentic traces, and per-knowledge-base metrics mean teams get the monitoring required for enterprise deployments without additional tooling.
- Security and Compliance: Document-level ACL enforcement at retrieval time and availability in AWS GovCloud (US-West) make the service suitable for regulated industries and sensitive enterprise data.
- Accelerates the RAG Adoption Curve: Optimized defaults for embeddings, chunking, and re-ranking remove weeks of experimentation typically needed to achieve production-quality retrieval accuracy.
- Ecosystem Lock-In Reduction: MCP-compatible integration means organizations are not forced to use a single agent framework, preserving architectural flexibility.
How It's Different
- Fully Managed vs. Self-Managed: Unlike the self-managed Knowledge Base option (where customers bring their own vector stores), Managed Knowledge Base owns the entire stack—storage, embeddings, re-ranking, and scaling—with no infrastructure to operate.
- Smart Parsing vs. Manual Configuration: Competitors and the self-managed path require developers to choose and tune chunking/parsing strategies per document type; Smart Parsing selects the optimal strategy automatically per document.
- Agentic Retrieval vs. Single-Shot RAG: Most RAG implementations perform a single retrieval pass; Agentic Retrieval iteratively plans, retrieves, evaluates, and re-ranks across multiple knowledge bases for complex multi-hop queries without custom orchestration.
- Native AgentCore/MCP Integration: The service exposes the knowledge base as an MCP-compatible tool automatically, whereas other RAG services require manual API wiring to agent frameworks.
- Six Pre-Built Enterprise Connectors: Native connectors for SharePoint, Confluence, Google Drive, and OneDrive are included out of the box, reducing the need for third-party ETL tools or custom ingestion code.
- Document-Level ACL Filtering at Retrieval Time: Fine-grained permission enforcement at query time (not just at ingestion) is a differentiator from many vector database solutions that require application-layer access control.
When to Prefer It
- Rapid Prototyping to Production: When a team needs to move from a RAG proof-of-concept to a production deployment quickly without hiring infrastructure specialists.
- Enterprise Knowledge Assistants: When building internal employee assistants that must query across SharePoint, Confluence, Google Drive, and OneDrive with per-user permission enforcement.
- Customer Support Automation: When automating customer-facing support that requires accurate, grounded responses from large, frequently updated document repositories.
- Complex Multi-Hop Queries: When user queries require reasoning across multiple documents or data sources (e.g., "What is our refund policy for enterprise customers in EMEA?"), making Agentic Retrieval more effective than single-pass RAG.
- Multimodal Knowledge Bases: When the enterprise data corpus includes images, audio recordings, video, or scanned documents that need to be retrievable alongside text.
- Regulated or Government Workloads: When data residency, compliance, or government security requirements mandate AWS GovCloud deployment with auditable access controls.
- Teams Without MLOps Expertise: When the organization lacks the expertise to manage vector database tuning, embedding model selection, or retrieval pipeline optimization.
Availability
- Status: Generally Available (GA) as of June 17, 2026.
- Supported Regions: US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney, Tokyo), Europe (Dublin, Frankfurt, London), and AWS GovCloud (US-West).
- Pricing Model: Not explicitly stated in the announcement; refer to the Amazon Bedrock Knowledge Bases pricing page for current rates (typically based on data stored, queries processed, and data ingested).
- Self-Managed Option Still Available: Teams requiring full pipeline control can still use the self-managed Knowledge Base path with their own vector stores.
- Web Crawler ACL Limitation: Document-level Access Control List filtering is not supported for the Web Crawler connector at this time.
- Model Defaults: AWS manages default embedding and re-ranking models, with the option for customers to bring their own models for customization.
