models
From code to diagrams: Agentic architecture documentation with Amazon Bedrock AgentCore
Amazon Bedrock AgentCore enables automated analysis and documentation of software architecture, streamlining the creation of up-to-date diagrams and searchable documentation through an integrated AWS pipeline, exemplified by a global financial services firm.
AS1 NewsSource: aws.amazon.com
AMZN
COSTMaintaining current and comprehensive architecture documentation remains a significant challenge in software development, especially as codebases evolve rapidly. Manual processes for creating diagrams are time-consuming and often become outdated quickly, leading to knowledge silos, onboarding delays, and compliance risks.
Amazon Bedrock AgentCore offers an autonomous solution that analyzes code bases, generates architecture diagrams, and maintains searchable documentation automatically. This approach leverages iterative refinement and self-correction, orchestrated through AWS services such as CodePipeline, S3, and Bedrock Knowledge Bases.
A global interdealer broker has implemented this system to keep architecture documentation current across their electronic trading platform. The pipeline triggers on code commits, analyzes .NET codebases with AgentCore, and generates diagrams stored in S3. These diagrams, along with metadata, are ingested into Bedrock Knowledge Bases, enabling semantic search and natural language queries.
The workflow involves code analysis, diagram generation, syntax validation, conversion to SVG, and uploading artifacts for indexing. The agent focuses on production code, prioritizing efficiency and relevance. It iteratively refines diagrams through self-correction, achieving high reliability.
This system integrates several AWS components: AgentCore for code analysis, CodePipeline for orchestration, S3 for storage, CodeBuild for execution, and Bedrock Knowledge Bases for search and retrieval. The process supports natural language queries, allowing users to explore architecture diagrams and their context easily.
The solution has been operational since December 2025, generating over 140 diagrams weekly across 20 repositories with minimal manual effort. It has significantly reduced documentation time from days to minutes, improved onboarding, and enhanced compliance readiness, demonstrating substantial business value.
Teams can start with pilot projects, expand to more repositories, and integrate with existing documentation tools. The approach offers a scalable, reliable, and cost-effective method to maintain current architecture documentation, supporting faster decision-making and organizational knowledge sharing.
The implementation improves architecture documentation processes, enhances compliance and onboarding, and reduces manual effort and errors in enterprise environments.