AWS Bedrock Managed Knowledge Base ships with 6 connectors

Key Takeaways

• AWS claims the new service cuts RAG pipeline engineering time by 90%
• Six native connectors support S3, SharePoint, Confluence, Google Drive, OneDrive, and web crawling
• Smart Parsing automatically selects chunking strategies per data type, removing manual optimization

Amazon Bedrock Managed Knowledge Base is now available, giving developers a single managed service to connect foundation models with enterprise data. AWS says the offering automates retrieval-augmented generation pipelines, handling storage, embeddings, re-ranking, and model selection so teams can skip the infrastructure work and go straight to building agents.

The pitch is straightforward: enterprises want generative AI grounded in their own documents, but the plumbing to get there, custom connectors, chunking logic, vector stores, and re-rankers, eats months of engineering time. AWS claims this new service cuts that effort by 90%.

What does Managed Knowledge Base actually manage?

At its core, the service abstracts five components developers normally wire together themselves: storage, retrieval, embeddings, re-ranking, and foundation model selection. By default, AWS picks and operates the embeddings model, re-ranker, and base LLM. Developers who want control can override these choices, but the point is they don't have to.

On top of that foundation, AWS highlights three features.

• Native data connectors: Six pre-built integrations for Amazon S3, SharePoint, Confluence, Web Crawler, Google Drive, and OneDrive. These pull data and permissions directly from each source.
• Smart Parsing: The system inspects content type and source, then applies an appropriate parsing and chunking strategy automatically.
• Agentic Retriever: Handles multi-turn, multi-hop queries that span multiple knowledge bases or require intent inference.

Swami Sivasubramanian, AWS VP of Database, Analytics, and AI, framed the value bluntly: "By abstracting away the heavy lifting of RAG infrastructure, we are enabling developers to focus entirely on building agents that deliver actual business value rather than managing complex data pipelines."

How do you create a knowledge base?

AWS positions setup as a matter of minutes. From the Bedrock AgentCore console or the standard Bedrock console, developers open the Knowledge Bases page and click Create Managed KB. A dropdown lists supported connectors; IAM roles generate automatically, though they can be edited.

The console presents optimized defaults. Once data syncs, the knowledge base plugs into agents via Bedrock AgentCore Gateway. Integration requires a few lines of code, and the AgentCore Observability dashboard surfaces evaluation metrics.

Why is chunking and parsing so hard?

RAG accuracy depends heavily on how documents are split before embedding. Chunk too large and retrieval returns irrelevant context. Chunk too small and you lose meaning. PDFs, HTML, markdown, and spreadsheets each need different treatment.

Historically, teams experimented with splitting strategies per data source, sometimes for weeks. Smart Parsing attempts to automate this experimentation. AWS has not published benchmark comparisons, but developer forums show interest in seeing how the automated approach stacks up against hand-tuned pipelines on domain-specific corpora.

What are developers saying?

Early reactions on Hacker News and Reddit lean positive. Developers praise the escape from "infrastructure fatigue" and the native support for SharePoint and Confluence, two sources notorious for complex authentication and metadata. The managed aspect, automatic scaling, patching, and model updates, removes ongoing maintenance overhead.

Skeptics raise two concerns. First, cost: managed services often charge premiums over self-hosted alternatives, and scaling to millions of documents may get expensive. AWS has not disclosed pricing details in the announcement. Second, control: teams with strict retrieval requirements want to know whether they can swap out embedding models, adjust re-ranking weights, or customize chunking logic beyond what Smart Parsing selects.

Where does this fit in AWS's AI stack?

Bedrock already offers foundation model access, guardrails, and agent orchestration. Managed Knowledge Base slots in as the data layer for agentic applications. It competes with LangChain-based self-assembly, Azure AI Search integrations, and Google's Vertex AI RAG Engine. AWS's bet is that most enterprises will trade some customization for faster deployment and lower maintenance.

The service supports near real-time synchronization, so agents see the latest enterprise data as it changes. For organizations where stale information is a compliance or accuracy risk, this matters.

Frequently Asked Questions

What data sources does Amazon Bedrock Managed Knowledge Base support?

At launch, six connectors: Amazon S3, SharePoint, Confluence, Google Drive, OneDrive, and Web Crawler.

Does Managed Knowledge Base handle document chunking automatically?

Yes. Smart Parsing inspects each data type and selects the parsing and chunking strategy automatically.

Can I customize the embedding or re-ranker models?

The service selects defaults, but AWS indicates developers can override these choices if needed.

How does Managed Knowledge Base integrate with Bedrock agents?

It's a pre-built target in Bedrock AgentCore Gateway. Integration requires a few lines of code, and observability metrics appear in the AgentCore dashboard.

Is pricing available for the new service?

AWS has not disclosed specific pricing in the announcement. Details are expected in the Bedrock pricing documentation.

Source: AWS News Blog