Diagram Agent
Takes user input in natural language to create architecture diagrams
DiagramAgent is an open-source, AI-powered tool that generates professional cloud architecture diagrams from natural language prompts, complete with product and service icons (AWS, Azure, GCP, Kubernetes, and more).
This is very much a work in progress. I built it because there’s currently no open-source or first-party Microsoft solution for generating technical architecture diagrams - the kind you’d typically create in Visio, draw.io, or Lucidchart, using AI with proper product and feature iconography. A few paid products from dedicated companies exist in this space, but nothing freely available.
Under the hood, it’s built on D2, an open-source diagram-as-code framework, paired with a Next.js frontend, Monaco code editor, and a WASM-based render engine. The workflow is conversational: describe your architecture, answer a few AI-generated clarifying questions to refine scope and detail, and watch the diagram stream in live. It also includes an experimental vision-based diagram scorer GPT-4o that evaluates the rendered output and iteratively feeds corrections back to the LLM to improve layout, completeness, and visual quality.
Turn your tech dreams into stunning cloud architecture visuals just by chatting like a pro! It’s like having a design buddy who speaks fluent AI and turns your words into sleek diagrams instantly.
Key features:
- 🗣️ Natural language input to create stylish architecture diagrams
- 🤖 AI-powered clarifying questions make sure your design is spot on
- 🎨 Vision refinement with iterative AI tweaks for picture-perfect output
- 📊 Export diagrams as SVG or PNG with live pan and zoom
This summary was generated by GitHub Copilot based on the project README.