next-ai-draw-io: Natural Language to draw.io Diagrams

Best For

Researchers, engineers, and technical writers who frequently create system architecture diagrams, data flow charts, or proposal illustrations. If you already use draw.io but spend too much time dragging boxes from scratch, this tool gives you a usable first draft from a single sentence.

How I Actually Use It

My typical workflow: describe the system in one or two sentences, let the AI generate a draw.io XML draft, then open the result in draw.io for precise alignment and styling. For simple diagrams (5-10 nodes), the AI output is often good enough after one round of conversational refinement. For complex diagrams, I treat the AI output as scaffolding and do manual layout adjustments.

I also use the image-to-diagram feature to convert whiteboard photos into editable draw.io files. Upload a photo, the AI reconstructs it as XML, and I clean up from there.

The MCP Server mode plugs into Claude Desktop, Cursor, or VS Code, so an AI agent can generate diagrams as part of a larger workflow without switching to a browser.

Where It Is Strong

• Native draw.io output: unlike Mermaid-based tools, the output is draw.io XML with full editing capability (precise positioning, custom styles, grouping, layers)
• Multi-provider support: works with 12+ LLM providers including OpenAI, Anthropic, Google, AWS Bedrock, and Ollama for fully offline use
• Self-hostable: Docker, Cloudflare Workers, or plain npm run dev. You control where your data goes
• Active maintenance: 30.9k stars, 675 commits, 18 releases, Apache 2.0 license

Where It Fails

• Complex diagrams break down: beyond roughly 15 nodes or with deeply nested structures, the AI-generated layout gets messy. You will need to manually rearrange
• Installation barrier: requires Node.js, npm, and at least one LLM API key (or a local Ollama setup). Non-technical users will struggle with setup
• Prompt describing visual layout is hard: getting the AI to place specific elements in specific positions requires verbose prompting. It understands logical relationships better than spatial arrangements

Pricing, Difficulty, and Risk

Pricing: fully open-source under Apache 2.0. Your only cost is the LLM API usage (or free if you run Ollama locally).

Difficulty: intermediate. Comfortable with npm install and .env files? You will be fine. If terminal commands feel foreign, this is not the right entry point.

Risk: API keys are stored in the browser (BYOK model) and the project claims no server-side persistence. However, all conversation content, including your diagram descriptions, is sent to whichever LLM provider you select. For sensitive or proprietary architecture diagrams, self-host with Ollama to keep everything local.

Verdict

If you use draw.io regularly and want AI to handle the tedious first-draft phase, next-ai-draw-io is the best open-source option available. Treat it as a drafting accelerator, not a finished-diagram generator. Skip it if you cannot run npm or if your diagrams never exceed what Mermaid can handle.

Source

• GitHub: https://github.com/DayuanJiang/next-ai-draw-io