Draw system diagrams from text and refine them via chat
Translating a mental system structure onto a canvas often takes time to place shapes and add labels. This tool draws diagrams on a draw.io canvas from natural language descriptions and lets you select specific shapes to modify them through chat. The output is editable like a standard draw.io file and can be exported in .drawio, .png, and .svg formats.
Generating diagrams from text and images
You can generate architecture diagrams, flowcharts, and sequence diagrams with a single sentence, including icon libraries for AWS, Azure, GCP, and Kubernetes. Uploading a screenshot or sketch allows the AI to redraw it, while uploading text files such as PDFs, Markdown, or code converts their content into a diagram. Generated diagrams function identically to standard draw.io files, allowing you to rename items by double-clicking, drag elements, and change styles. You can also apply flowing animation effects to connection lines.
Edit history and selective editing
Each time the AI modifies a diagram, it records a version card, allowing you to compare thumbnails or revert to a previous state. Pressing Ctrl+Z on the canvas undoes the AI's last modification in one step. If you instruct the AI while specific shapes are selected, it modifies only those selected parts. Modified shapes are highlighted on the canvas to make changes easy to verify.
AI agent integration and deployment methods
The tool is available as a web app, Windows, macOS, and Linux desktop apps, and an MCP server. Through the MCP server, AI agents such as Claude Code, Cursor, and VS Code can directly generate and modify diagrams. API keys are stored in the browser and not sent to the server, with support for more than 25 model providers. Local model usage via Ollama is possible, and you can deploy to your own server using Vercel, Cloudflare Workers, Docker, and others.
Model selection and pre-deployment checks
Because this tool generates long text in draw.io XML format, it is recommended to use a high-performance model rather than a small one. The free demo site has usage limits, so you must enter your own API key to use it without restrictions. Self-deployment requires setting environment variables and specifying model IDs, and you can manage models and access codes through an admin panel. License information is not specified in the README, so you must directly check the license file in the repository before applying it to commercial projects.
