JA EN
shiichan

Say goodbye to fiddly RAG setup! Amazon Bedrock Managed Knowledge Base is here!

AWS Blog ★★★★ View Markdown
#aws#ai

Hi everyone, it's Shii-chan! Today I have great news for those of you building generative AI for the enterprise. The RAG pipeline work that used to be such a pain can now be set up in just minutes with a brand-new service!

AWS Blog aws.amazon.com

What was announced?

This one comes from the official AWS News Blog (AWS Blog): Amazon Bedrock Managed Knowledge Base. It's a new service that lets you build enterprise-grade generative AI applications on your own proprietary data in minutes. It takes the retrieval-augmented generation (RAG) pipeline that used to be so much work to build and operate, and makes the whole thing fully managed.

The story so far

RAG is wonderful, but building it yourself turned out to be surprisingly hard. The post lists three walls people tend to hit.

  • Connecting data spread across different systems, each with its own access controls and formats
  • Keeping up with RAG accuracy best practices that keep evolving over time
  • Operating at scale across millions of documents, or thousands of smaller knowledge bases

Handling all of that yourself was quite a lot of work.

What changes

With Managed Knowledge Base, the service takes on those tedious parts for you. Developers open the Amazon Bedrock or AgentCore console, choose "Create Managed KB," connect a data source from a dropdown, and that's it. You get optimized defaults, and once your data syncs you can start querying right away. The post says you can get going with "just a few lines of code."

The nice part is that you can focus on business outcomes instead of managing infrastructure.

Let's dive deeper

Let's look at what's inside a bit more closely.

Native data connectors

Six ready-to-use connectors pull enterprise data in directly.

  • Amazon S3
  • SharePoint
  • Confluence
  • Web Crawler
  • Google Drive
  • OneDrive

Smart Parsing

This automatically picks the best parsing strategy for each data type and connector.

  • Connector-specific data models preserve document hierarchy and relationships
  • Multimodal processing detects and extracts images, bounding boxes, and video content
  • Optimized chunking uses foundation models to understand document structure

Agentic Retriever

This is a retrieval system built for complex, multi-step queries. It performs multihop retrieval across a single knowledge base or several at once. It decomposes a query into a step-by-step plan, runs intermediate evaluations along the way, and stops searching once it has gathered enough relevant passages.

AgentCore Gateway integration

Managed Knowledge Base works as a native pre-built target type within AgentCore Gateway. It auto-generates IAM role-based permissions and gives you visibility through the AgentCore Observability dashboard. It also implements the Model Context Protocol (MCP), so frameworks like LangChain, CrewAI, LlamaIndex, and LangGraph can discover it automatically.

Model flexibility, regions, and pricing

Every Bedrock foundation model can power generation, and you can pick from different embedding and re-ranking models. It supports specialized embedding models for domain-specific needs, and it's backward compatible with existing Bedrock Knowledge Base APIs.

Available regions are US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney, Tokyo), Europe (Dublin, Frankfurt, London), and AWS GovCloud (US-West). It's nice that Tokyo is included.

Pricing is pay-per-use with no upfront commitments, based on the size of indexed data stored and the number of retrievals performed (on-demand). New customers can also use it under the AWS Free Tier.

Wrap-up

Here's a quick recap.

  • Amazon Bedrock Managed Knowledge Base makes building and running RAG pipelines fully managed
  • Six native connectors (S3, SharePoint, Confluence, Web Crawler, Google Drive, OneDrive) pull data in directly
  • Smart Parsing automates multimodal-aware parsing, and the Agentic Retriever handles multi-step search
  • AgentCore Gateway integration adds MCP support, usable from major agent frameworks like LangChain
  • Pay-per-use pricing, Free Tier for new customers, and it's available in the Tokyo region too

This is a perfect fit for enterprise developers who want to spin up generative AI apps on their own data quickly, without wrestling with infrastructure!

Related posts