Qwen OpenAI Proxy

An OpenAI-compatible proxy server that routes requests from any OpenAI SDK client to Qwen models via chat.qwen.ai.

• Drop-in replacement for api.openai.com
• Streaming SSE support
• OpenAI tool calling compatibility
• Session persistence
• System prompt merging for Qwen compatibility

Qwen provides powerful models through chat.qwen.ai, but does not expose a fully OpenAI-compatible API.

This proxy allows existing OpenAI SDKs, coding agents, and developer tools to use Qwen models without modification.

• OpenAI-compatible API
• Streaming responses (SSE)
• Tool/function calling
• Conversation persistence
• Automatic system message merging
• Model aliases
• OpenCode support
• LangChain support
• Vision uploads
• Docker image
• Multi-account rotation

Works with:

• OpenCode
• Cline
• Continue.dev
• LangChain
• OpenAI SDK
• Roo Code
• OpenWebUI
• LibreChat
• AnythingLLM
• Cursor

Any OpenAI-compatible client should work.

• Node.js 18+
• npm or pnpm

npm install
npm run build
npm start

Server runs at:

http://localhost:5000/v1

npm run login

This opens a browser window.

1. Log in to chat.qwen.ai
2. Press Enter in the terminal
3. Tokens and cookies are extracted automatically

1. Log in to https://chat.qwen.ai
2. Open DevTools (F12)
3. Go to Console tab
4. Run:

localStorage.getItem("token")

5. Copy the token
6. Open Network tab
7. Copy the Cookie header from any request
8. Create .env:

QWEN_AUTH_TOKEN=your_token_here
QWEN_COOKIE="your_cookie_here"

curl http://localhost:5000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model":"qwen-coder",
    "messages":[
      {
        "role":"user",
        "content":"Write a Python web scraper"
      }
    ]
  }'

import OpenAI from "openai"

const client = new OpenAI({
  baseURL: "http://localhost:5000/v1",
  apiKey: "dummy"
})

const response = await client.chat.completions.create({
  model: "qwen-coder",
  messages: [
    {
      role: "user",
      content: "Hello"
    }
  ]
})

console.log(response.choices[0].message)

Add this to your opencode.json providers:

"localproxy": {
  "npm": "@ai-sdk/openai-compatible",
  "name": "Local Proxy",
  "options": {
    "baseURL": "http://localhost:5000/v1",
    "apiKey": "dummy"
  },
  "models": {
    "qwen3.7-max": {
      "name": "Qwen 3.7 Max"
    },
    "qwen3.6-plus": {
      "name": "Qwen 3.6 Plus"
    }
    // Add any model from /v1/models above
  }
}

{
  "model": "qwen-coder",
  "messages": [
    {
      "role": "user",
      "content": "Read config.json"
    }
  ],
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "read_file",
        "description": "Read a file",
        "parameters": {
          "type": "object",
          "properties": {
            "path": {
              "type": "string"
            }
          },
          "required": ["path"]
        }
      }
    }
  ]
}

Direct upstream model IDs are also supported.

OpenAI-compatible chat completions endpoint.

Lists available models.

curl http://localhost:5000/v1/models

Deletes stored chat sessions.

Refreshes Qwen authentication credentials.

Client (OpenAI SDK / OpenCode / LangChain)
                ↓
        Qwen OpenAI Proxy
                ↓
          chat.qwen.ai
                ↓
           Qwen Models

src/
├── index.ts
├── auth/
├── client/
├── proxy/
├── streaming/
├── tools/
├── session/
└── utils/

Qwen enforces a strict constraint:

• Maximum one system message
• System message must be first

The proxy automatically:

• merges multiple system prompts
• preserves tool messages
• preserves assistant messages
• fixes compatibility with OpenCode and similar agents

# Build
docker build -t qwenchat2api .

# Run
docker run -d \
  -p 5000:5000 \
  -e QWEN_AUTH_TOKEN=your_token \
  -e QWEN_COOKIE="your_cookie" \
  -v qwenchat2api-data:/app/data \
  --name qwenchat2api \
  qwenchat2api

• chat.qwen.ai is not an official public API
• This project reverse-engineers the web interface
• Tokens expire periodically
• Rate limits are undocumented and account-dependent
• Some upstream model names may change

• Vision/image uploads
• Docker image
• WebSocket support
• Multi-account routing
• Better retry handling
• OAuth authentication flow

This project is unofficial and is not affiliated with Alibaba or Qwen.

Use responsibly and at your own risk.

MIT