AWS Builder Center
OpenArchFlow: Generate AWS Architecture Diagrams from Natural Language Using AI

OpenArchFlow: Generate AWS Architecture Diagrams from Natural Language Using AI

A hands-on look at OpenArchFlow, a free open-source PWA that transforms plain English descriptions into interactive AWS architecture diagrams using Google Gemini or fully local AI — no account, no setup, no data leaving your browser.

Software Engineer
If you've ever spent hours fighting with diagram tools — dragging boxes, hunting for the right AWS icon, trying to remember if the arrow goes left or right — OpenArchFlow was built for you.

What Is OpenArchFlow?

The project lives at  openarchflow.cloud  and the source code is publicly available at  github.com/dmux/OpenArchFlow **** under the MIT license. No sign-up, no paid tier, no vendor lock-in.

The Problem It Solves

reating architecture diagrams is tedious. Traditional tools require you to manually place every icon, draw every arrow, and label every connection. If you change your mind about the design, you start over. If you want to share it with non-technical stakeholders, you export a static image and immediately lose the interactivity.
OpenArchFlow flips this workflow. Instead of building a diagram and then describing it, you describe what you want to build and the diagram appears. The AI understands AWS service relationships, automatically labels data-flow connections, and lays everything out hierarchically. You can then drag, drop, and refine from there.

Key Features at a Glance

Switch between Cloud AI (Google Gemini 2.5 Flash) for speed and Local AI (Phi-3-mini via WebLLM/WebGPU) for complete privacy. The local model runs 100% in your browser with no internet required.
  • Incremental Generation: Already have a diagram? Just describe what you want to add or change and the AI modifies the existing diagram without starting from scratch. This is great for evolving architectures during a design session.
  • Diagram Chat: Use the built-in AI assistant to ask questions directly about your architecture: "What would this cost?", "Are there any security gaps?", or "How would I add a caching layer?" The assistant responds in context.
  • Architecture Simulation: Animate data flows to visualize traffic patterns through your components. Perfect for presentations and walkthroughs with your team.
  • AI-Generated Technical Documentation: One click generates a full Markdown specification document including executive summary, component descriptions, key configurations, and best practices for security and scalability.
  • Massive Component Library: Hundreds of official icons for AWS, Azure, Cloud Native, Datadog, Sentry, Stripe, Twilio and generic shapes. Everything is drag-and-droppable from a searchable sidebar.
  • Privacy-First Architecture: No data collection, no analytics, no cookies, no telemetry. Your diagrams are stored locally in your browser's localStorage/IndexedDB. Not even the API key is sent anywhere except directly to the AI provider.

Hands-On: Generating a Serverless E-Commerce Architecture

No account needed, just open the browser and start. Here's what happened when I typed the following prompt into the AI input bar:
"Serverless e-commerce backend with API Gateway, Lambda, DynamoDB, S3, Cognito, and CloudFront"
Within seconds, the AI (Google Gemini 2.5 Flash) generated a complete, hierarchical diagram with all the requested AWS services — correctly connected with labeled edges like "forwards API requests to", "authenticates with", "stores/retrieves product images", and "emits logs & metrics". Every node displayed the proper official AWS icon.

AI-Generated Architecture Documentation

Under the Hood: Tech Stack

A library that runs Phi-3-mini entirely in the browser via WebGPU, meaning the model weights are downloaded once and then the inference happens locally with no network calls at all. This is a genuinely privacy-preserving setup that goes beyond what most "local AI" products actually deliver.

How to Get Started

Why This Matters for the AWS Community

Used in design reviews, customer presentations, RFCs, and incident post-mortems. But the tooling has lagged behind. OpenArchFlow represents a meaningful step forward in making diagram creation as fast as thinking.
What sets it apart from other AI diagramming experiments is the combination of practical AWS depth (it knows service relationships and recommended patterns), zero-friction access (open the URL and start), and a genuine commitment to privacy (the local AI mode is not a marketing claim — it actually works offline). The multi-diagram management, backup/restore, and versioned JSON export mean it is usable as a real day-to-day tool, not just a demo.
There are areas to grow. Mobile support is still limited, the local Phi-3 model is less capable than Gemini for complex multi-service architectures, and the component library could benefit from more Azure and GCP completeness. But these are known areas the project is actively developing, with community contributions welcome.

Try It Today

The roadmap includes more AWS service coverage, internationalization, mobile support, and improved AI prompt engineering. It is a genuinely useful open-source contribution to our community's toolkit.

Links & Resources

The diagram included: CloudFront (CDN), API Gateway, Lambda (API handler), Amazon Cognito (user auth), SQS (async queue), Lambda (SQS processor), S3 (static assets), DynamoDB (data store), and CloudWatch (monitoring). All organized in a clean top-to-bottom flow that would take 30+ minutes to build manually in a traditional diagramming tool.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article