Building document-investigation agents with Amazon Bedrock AgentCore and Amazon S3 Files

This article demonstrates how to build document-investigation agents using Amazon Bedrock AgentCore and Amazon S3 Files, enabling agents to autonomously discover and analyze related documents through a shared file system.

• Lambda pre-stages documents to Amazon S3 Files; AgentCore agent mounts same file system to investigate them
• Agent reads documents, follows cross-references, and writes findings using ordinary file operations without S3 API calls
• S3 Files provides NFS-based shared access with read-after-write consistency and POSIX permissions
• Maintain consistency by reading from mount, assigning one writer per path, and naming files per producer
• Mitigate prompt injection risks by validating case IDs, separating write paths, requiring evidence, and screening documents with Guardrails
• Reference implementation available on Serverless Land with CloudFormation template and agent tools

This serverless approach eliminates download/upload orchestration overhead while enabling intelligent, autonomous document investigation workflows across multiple related files.