Dayuan Jiang
99ced88e5b
## Problem
A review of how the app uses the Vercel AI SDK (two Claude reviewers and Codex, each finding checked against the SDK's type definitions and source) found five behaviours that are wrong and a set of places where the app re-implements what the SDK already provides:
1. Quota tokens were recorded in `onFinish`, which the SDK does not call when a later step fails; only a user stop was handled (`onAbort`), and the write was not awaited.
2. An automatic retry after a tool error re-sent the request body as it was when the message was sent, so the model was told the "authoritative" canvas XML was the diagram from before the turn, although `display_diagram` commits a drawing before its screenshot check can fail.
3. The screenshot check and the connection test ran their model call on `AbortSignal.timeout` only, so pressing Stop (or closing the dialog) did not stop a paid model call.
4. The client ignored `finishReason`; a reasoning model that spent the whole output budget before calling a tool left the user with no diagram and no message.
5. The connection test counted an `invalid` tool call (a tool that does not exist, unreadable input) as proof of tool support.
## Change
- `onStepFinish` records each step's `usage.totalTokens` and awaits the write; the copies in `onFinish` and `onAbort` are gone.
- `DefaultChatTransport.prepareSendMessagesRequest` sends the current canvas page XML on every request after the first of a turn; the first keeps the exported (or snapshot) XML.
- `validate-diagram` and the connection test pass `req.signal` and use the SDK's `timeout` option; `validate-diagram` reports to Langfuse under `functionId: "validate-diagram"`.
- The connection test only counts a valid tool call.
- The client's `onFinish` shows a notice (with Retry) when `finishReason === "length"` and no tool was called.
- The server validates incoming messages with `safeValidateUIMessages` (400 otherwise), drops client system messages explicitly and prunes empty model messages with `pruneMessages`.
- The chat's tools live in `lib/chat-tools.ts` with an `outputSchema` for the client-side ones; `lib/chat-message.ts` defines `ChatUIMessage` (metadata, a `data-notice` part, `InferUITools`) so `useChat`, `addToolOutput` and the tool handlers are typed.
- Error notices are assistant messages with a `data-notice` part instead of role `system` UI messages.
- Regenerate, edit and retry use `sendMessage({ messageId })`; `isResending` and `flushSync` are gone.
- `isToolUIPart` / `getToolName` replace slicing `"tool-"` off part types; `ToolPartLike` derives from the SDK's part types.
Net: 578 insertions, 439 deletions over 16 files.
## Testing
- New unit tests: a step that fails after a finished step still counts that step's tokens; the screenshot check's model call ends when the request is aborted; an invalid tool call in the connection test gives the warning; malformed messages get a 400. The first and third confirm SDK behaviour the review relied on.
- `tsc`, `biome ci` (0 errors), root `vitest run` 723 passed on the branch rebased onto #986, `next build`: clean
- Playwright, 10 chat-related specs on the dev server: 199 passed, 0 failed, both before and after the rebase onto #986
Next AI Draw.io
Draw and edit draw.io diagrams by chatting with an AI.
https://github.com/user-attachments/assets/66b9f12f-219f-4d62-acc0-0725e6850eec
Describe a diagram in a sentence and the AI draws it on a real draw.io canvas. Edit it by hand like any draw.io file, or select a few shapes and tell the AI what to change. Every AI change is a version you can compare, restore or undo, and the result exports as .drawio, .png or .svg.
Available as a web app, a desktop app for Windows, macOS and Linux, and an MCP server for AI agents such as Claude Code, Cursor and VS Code.
Highlights
Draw
- Architecture diagrams, flowcharts, sequence diagrams and more from one sentence, with built-in AWS, Azure, GCP and Kubernetes icon libraries
- Connectors can carry a flowing animation
- Upload a screenshot or a sketch and the AI redraws it; upload PDF, Markdown, code and other text files and get a diagram of their content
Edit
- Edit by chat: changes stream onto the canvas and the shapes the AI just changed are highlighted
- Ask about a selection: select shapes on the canvas and the AI changes only those
- Versions and undo: every AI change is a version card with a thumbnail; compare it with the canvas, restore it or undo it, and Ctrl+Z on the canvas also takes back an AI change in one step
- It is a normal draw.io diagram: double-click to rename, drag, restyle, use several pages, and export as
.drawio,.png,.svgor.drawio.svgat any time
Use
- 24 model providers; enter your own API key in the browser and it stays on your machine
- Models that reason show their thinking
- Dark mode; the interface is available in English, Simplified Chinese, Traditional Chinese and Japanese
Examples
|
Transformer architecture with animated connectors Prompt: Give me a animated connector diagram of transformer's architecture. 
|
|
|
RAG architecture Prompt: Generate a RAG architecture diagram for chat application. Use connected diagram for data ingestion 
|
Authentication with React and AWS Prompt: Generate authentication process using React with AWS. Use Serverless architecture. 
|
|
Open Innovation model Prompt: Create visualization of Henry Chesbrough's Open Innovation model. 
|
Cat sketch Prompt: Draw a cute cat for me. 
|
Use it
Online demo
Open next-ai-drawio.jiang.jp, nothing to install. The demo has a usage limit; click the settings icon in the chat panel and enter your own provider and API key to lift it. The key stays in your browser and is never sent to the server for storage.
Desktop app
Download the Windows, macOS or Linux installer from the Releases page.
In your AI agent (MCP)
Through MCP (Model Context Protocol, the protocol AI agents use to call outside tools), Claude Desktop, Cursor, VS Code and others can draw draw.io diagrams directly. Add this to your client's MCP configuration:
{
"mcpServers": {
"drawio": {
"command": "npx",
"args": ["@next-ai-drawio/mcp-server@latest"]
}
}
}
For Claude Code, one command does it:
claude mcp add drawio -- npx @next-ai-drawio/mcp-server@latest
Then ask the AI for "a flowchart of user authentication with login, MFA and session management" and the diagram appears in your browser as it is drawn. The MCP server has most of the web app's drawing features:
- The same drawing rules and icon libraries (AWS, Azure, GCP, Kubernetes and more)
- The preview is the same canvas as the web app: the AI's changes are outlined, one Ctrl+Z takes them back, and the AI can read the shapes you select
- The draw.io editor ships inside the package, so after install the preview works offline (
DRAWIO_BASE_URLpoints it at an external draw.io instead) - A screenshot tool, so the AI can look at the rendered diagram and fix it
- Version history, multi-page diagrams, and download as
.drawio,.png,.svgor.drawio.svg - Auto-save to
~/.next-ai-drawio/, so a diagram survives a restart - Ask the AI to undo, and keep your own drawing rules in
~/.next-ai-drawio/instructions.md
See the MCP server README for VS Code, Cursor and other client configurations.
Self-host
Run locally
git clone https://github.com/DayuanJiang/next-ai-draw-io
cd next-ai-draw-io
npm install
cp env.example .env.local # add your provider and API key, see "Models and providers" below
npm run dev
Open http://localhost:6002.
One-click deploy
| Platform | How |
|---|---|
| Tencent EdgeOne Pages |  Deploying there also gives you a daily free quota for DeepSeek models |
| Vercel | .env.local |
| Cloudflare Workers | Cloudflare deploy guide |
| Docker | Docker guide |
| Offline or intranet | Offline deployment |
Models and providers
24 providers are supported: AWS Bedrock (default), OpenAI, Anthropic, Google AI, Google Vertex AI, Azure OpenAI, Ollama, OpenRouter, AIHubMix, DeepSeek, SiliconFlow, SGLang, Vercel AI Gateway, Tencent EdgeOne, ByteDance Doubao, ModelScope, Zhipu GLM, Qwen, Qiniu, Kimi, MiniMax, Novita, Xiaomi MiMo and Atlas Cloud. The environment variables and notes for each are in the provider configuration guide.
Which model: the task is to produce long text in a strict format (draw.io XML), so pick a capable model; small models tend to produce broken diagrams.
Several models and the admin panel: list several model IDs in AI_MODEL separated by commas, or configure models from several providers with the AI_MODELS_CONFIG environment variable or an ai-models.json file, and every user can use them without a key of their own. Set ADMIN_PASSWORD and open /admin to manage models, access codes, feature switches, observability and quotas from a web page; see the admin panel guide.
Support
- Questions and ideas: open a GitHub issue or email me[at]jiang.jp
- Common problems: FAQ
- If the project is useful to you, consider sponsoring to help keep the demo site running