Smarter Knowledge Bases for Smarter AI Agents

We’re rolling out new features to the GenAI Platform that make it easier to build, manage, and improve the knowledge bases behind your AI agents. With web crawling, custom crawling rules, and one-click reindexing, you can keep your agents up to date with relevant, real-world information, without manual data collection or external storage. Combined with recent enhancements to our Retrieval-Augmented Generation (RAG) system, these updates can help your AI agents deliver faster, more accurate, and more context-aware responses from richer, better-organized data sources.

Web crawling for knowledge bases

The quality of your AI agent output depends on the data it can access. With the new web crawling feature, you can crawl publicly available websites and index content into your knowledge base, reducing the need for manual data collection. This is especially valuable for AI agents that rely on public web content to drive insights and actions.

• Custom crawling rules let you target specific pages or entire domains, ensuring your agent pulls data from the sites you tell it to.
• One-click reindexing keeps your knowledge base fresh and up to date, making it ideal for use cases like financial analysis and competitive research.
• No need for external storage, since crawled content is directly integrated into your knowledge base, simplifying setup and reducing overhead.

With web crawling, your AI agent can stay informed by pulling in real-time information.

RAG Improvements

To help your agents make the most of that knowledge, we’ve significantly upgraded our Retrieval-Augmented Generation (RAG) system.

• Accuracy on text-based questions has nearly doubled, reaching up to 95 percent accuracy*.
• Answers involving tables and graphs are now nearly four times more accurate*, powered by improved layout models and GPU-accelerated OCR.
• Metadata-aware responses improve context and clarity, allowing agents to reference details like source URLs, file names, and PDF page numbers.
• Metadata-based queries let users filter results directly through natural language, such as requesting a summary of a specific document.

These RAG enhancements help make your AI agents not just smarter, but also more precise, transparent, and useful across a wide range of real-world applications.

Bonus: new Anthropic model available—Claude 3.7

We’ve also added Claude 3.7 Sonnet to the GenAI Platform, giving you access to Anthropic’s latest and most advanced reasoning-focused model. With an extended thinking mode for deeper analysis and more accurate answers to complex queries, Claude 3.7 builds on the strengths of the Claude 3 family with improved performance, natural language fluency, and enhanced reliability. It’s ideal for agents that require strong problem-solving, advanced comprehension, and trusted responses.

Unlocking more possibilities for AI developers

Take your AI agents to the next level with our latest features, including real-time web crawling, enhanced RAG capabilities, and the powerful Claude 3.7 Sonnet. Try the new updates on the GenAI Platform today and elevate your AI-driven projects ->

*Internal benchmark study conducted in Q1 2025 across representative agent-building workloads using AWS Bedrock and leading alternatives. Time-to-build was measured for creating an Agent with Knowledge Base. Accuracy was evaluated using independent public data sets with domain-specific tasks across text, tabular, graphical, and multimodal data. Performance may vary, full methodology available upon request.