Qwen Worker Proxy

OpenAI-compatible API for Qwen's models (qwen3-coder-plus, qwen3-coder-flash, vision-model) deployed on Cloudflare Workers with multi-account support.

Overview

This project provides a Cloudflare Worker that acts as an OpenAI-compatible proxy for Qwen's AI models from free 2000 req/day of QwenLM/qwen-code. It handles OAuth2 authentication, token management, and provides standard OpenAI API endpoints.

Multi-account support allows you to manage multiple Qwen accounts for higher throughput and automatic failover.

Features

• ✅ OpenAI-compatible API endpoints
• ✅ Multi-account management with automatic rotation
• ✅ Automatic daily reset of failed accounts (no cron needed!)
• ✅ Intelligent account selection based on token freshness
• ✅ Automatic failover on quota exhaustion or errors
• ✅ OAuth2 authentication with automatic token refresh
• ✅ Global edge deployment via Cloudflare Workers
• ✅ Multiple models: qwen3-coder-plus, qwen3-coder-flash, vision-model
• ✅ Streaming support
• ✅ Token usage tracking
• ✅ KV-based token caching
• ✅ Admin health check endpoint

Prerequisites

1. Cloudflare Account with Workers enabled
2. Qwen OAuth Credentials from your existing authentication
3. Node.js and npm installed

Quick Start (Multi-Account)

Step 1: Set up Cloudflare Workers

1. Install dependencies:

   npm install

2. Authenticate with Cloudflare:

   wrangler login

Step 2: Add Qwen Accounts

Add one or more Qwen accounts with QR code authentication:

# Add first account
npm run auth:add account1

# Add second account
npm run auth:add account2

# Add more accounts...
npm run auth:add account3

# List all accounts
npm run auth:list

What happens:

• Shows QR code for authentication
• Auto-opens browser
• Saves credentials to ./.qwen/oauth_creds_{accountId}.json

Step 3: Deploy Accounts to KV

Deploy your accounts to Cloudflare KV storage:

# Deploy all accounts at once
npm run setup:deploy-all

# Or deploy individually
npm run setup:deploy account1

# List accounts in KV
npm run setup:list-kv

Step 4: Set Environment Variables

Easy way - Using .env file:

1. Create and edit .env file:

   cp .env.example .env
   nano .env  # Edit with your values

2. Update all secrets at once:

   npm run secrets:update

Manual way:

1. Set API keys for authentication (comma-separated for multiple keys):

   wrangler secret put OPENAI_API_KEYS
   # Enter: sk-key1,sk-key2,sk-key3

2. Set admin secret for health check endpoint:

   wrangler secret put ADMIN_SECRET_KEY
   # Enter your admin secret key

Verify secrets:

npm run secrets:list

Step 5: Deploy Worker

npm run deploy

Your multi-account proxy is now live! 🎉

Testing

Health Check

curl https://your-worker.workers.dev/health

List Models

curl https://your-worker.workers.dev/v1/models

Chat Completion

curl -X POST https://your-worker.workers.dev/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-coder-plus",
    "messages": [{"role": "user", "content": "Write hello world in Python"}]
  }'

Using New Models

# Using qwen3-coder-flash
curl -X POST https://your-worker.workers.dev/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-coder-flash",
    "messages": [{"role": "user", "content": "Write hello world in Python"}]
  }'

# Using vision-model
curl -X POST https://your-worker.workers.dev/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "vision-model",
    "messages": [{"role": "user", "content": "Describe this image if you can see it"}]
  }'

With API Key Authentication

curl -X POST https://your-worker.workers.dev/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-your-api-key" \
  -d '{
    "model": "qwen3-coder-plus",
    "messages": [{"role": "user", "content": "Explain recursion"}]
  }'

Using Python OpenAI SDK

from openai import OpenAI

client = OpenAI(
    base_url='https://your-worker.workers.dev/v1',
    api_key='sk-your-api-key-here'
)

# Using qwen3-coder-plus
response = client.chat.completions.create(
    model='qwen3-coder-plus',
    messages=[
        {'role': 'user', 'content': 'Explain recursion'}
    ]
)

print(response.choices[0].message.content)

# Using qwen3-coder-flash
response_flash = client.chat.completions.create(
    model='qwen3-coder-flash',
    messages=[
        {'role': 'user', 'content': 'Explain recursion'}
    ]
)

print(response_flash.choices[0].message.content)

# Using vision-model
response_vision = client.chat.completions.create(
    model='vision-model',
    messages=[
        {'role': 'user', 'content': 'Describe this image if you can see it'}
    ]
)

print(response_vision.choices[0].message.content)

Local Development

See LOCAL_DEVELOPMENT.md for detailed local setup instructions.

Multi-Account Management

Account Commands

# Authentication
npm run auth:add <account-id>     # Add new account with OAuth
npm run auth:list                 # List all local accounts
npm run auth:remove <account-id>  # Remove account credentials

# Deployment
npm run setup:deploy <account-id> # Deploy single account to KV
npm run setup:deploy-all          # Deploy all accounts to KV
npm run setup:list-kv             # List accounts in KV
npm run setup:remove-kv <account-id> # Remove account from KV

# Health Check
npm run setup:health              # Check status of all accounts

# Secrets Management
npm run secrets:update            # Update all secrets from .env file
npm run secrets:list              # List all configured secrets

Admin Health Check

Monitor all accounts via API endpoint:

curl https://your-worker.workers.dev/admin/health \
  -H "Authorization: Bearer your-admin-secret-key"

Response:

{
  "summary": {
    "total_accounts": 5,
    "healthy_accounts": 3,
    "failed_accounts": 2,
    "quota_exceeded_accounts": 1
  },
  "accounts": [
    {
      "account": "account1",
      "status": "healthy",
      "expiresIn": "45 min"
    },
    {
      "account": "account2",
      "status": "quota_exceeded",
      "expiresIn": "12 min"
    }
  ]
}

How Multi-Account Works

1. Automatic Account Selection: Probability-based selection favors accounts with freshest tokens
2. Automatic Failover: If one account fails, automatically tries another
3. Daily Reset: Failed accounts automatically reset at UTC midnight (lazy check, no cron needed)
4. Manual Cleanup: Permanently dead accounts must be removed manually via setup:remove-kv

Account Lifecycle

Quota Exhausted (429):

• Added to FAILED_ACCOUNTS list
• Auto-reset at UTC midnight
• Available again next day

Dead Account (Invalid Token):

• Added to FAILED_ACCOUNTS list
• Auto-reset at UTC midnight
• Fails again on next use
• Admin removes with: npm run setup:remove-kv <account-id>

API Endpoints

Endpoint	Method	Description
/health	GET	Health check
/admin/health	GET	Multi-account health status (requires admin key)
/v1/models	GET	List available models
/v1/chat/completions	POST	Create chat completion
/v1/debug/token	GET	Token info (dev only)

Environment Variables

Variable	Required	Description
OPENAI_API_KEYS	❌	Comma-separated API keys for authentication
ADMIN_SECRET_KEY	❌	Admin key for health check endpoint
OPENAI_API_KEY	❌	Single API key (legacy, deprecated)

Troubleshooting

Common Issues

• KV Namespace Not Found: Ensure you've created the KV namespace and updated wrangler.toml
• OAuth Credentials Invalid: Verify your credentials are properly formatted and not expired
• Deployment Fails: Check Cloudflare account permissions and Wrangler authentication

Debug Commands

For local development, see the debug tools in LOCAL_DEVELOPMENT.md.