Cloudflare AI Proxy

A Cloudflare Worker that provides OpenAI and Claude compatible API endpoints, powered by Cloudflare AI Gateway.

Features

• 🔄 OpenAI API Compatibility: /chat/completions, /models
• 🤖 Claude API Compatibility: /v1/messages
• 🔐 Dual Authentication: Bearer token and URL-based auth (JetBrains compatible)
• 📊 Request Logging: Built-in request/response logging

Local Development

Prerequisites

• Node.js
• pnpm (or npm)
• Cloudflare account

Setup

1. Clone the repository

   git clone <your-repo>
   cd cloudflare-ai-proxy

2. Install dependencies

   pnpm install

3. Create .dev.vars file

   The configuration uses JSON environment variables:

   cat > .dev.vars << 'EOF'
   CF_GATEWAY_KEY=your-cloudflare-gateway-key
   PROXY_API_KEY=your-secret-token
   MODELS_CONFIG=[{"id":"google-ai-studio/gemini-flash-latest","name":"gemini-flash-latest","endpoint":"/v1/YOUR_ID/proxy/compat"}]
   EOF

   Configuration Details:

   • MODELS_CONFIG: JSON array of model configurations

     [
       {
         "id": "google-ai-studio/gemini-flash-latest",
         "name": "gemini-flash-latest",
         "endpoint": "/v1/YOUR_ID/proxy/compat"
       }
     ]

   Replace YOUR_ID with your actual Cloudflare AI Gateway ID.

4. Start development server

   pnpm run dev

   The server will start at http://localhost:8787

Deployment to Cloudflare

Prerequisites

Before deploying, you need to configure model connections in Cloudflare AI Gateway:

1. Go to Cloudflare Dashboard

   • Navigate to AI → AI Gateway
   • Create a new Gateway or select an existing one
   • Note your Gateway ID (you'll need this for MODELS_CONFIG)

2. Configure Model Providers

   • Add your model providers (Google AI Studio, OpenRouter, etc.)
   • Set up API keys for each provider
   • Test the connections to ensure they work

Configuration

All environment variables should be configured on Cloudflare Dashboard (recommended).

Cloudflare Dashboard

1. Deploy project to worker

   • 

2. Go to Worker Settings

   • Navigate to Workers & Pages → Select your Worker → Settings → Variables
   • Select the "Production" environment

3. Configure Secrets (Encrypted)

   Add the following as Encrypted variables:

   • PROXY_API_KEY: Your custom authentication token for API access
   • CF_GATEWAY_KEY: Your Cloudflare AI Gateway API key

4. Configure Model Mapping (Plaintext)

   Add the following as Plaintext variable:

   • MODELS_CONFIG: JSON array mapping your models to AI Gateway endpoints

   Format:

   [
    {
      "id": "google-ai-studio/gemini-2.5-flash",
      "name": "gemini-2.5-flash",
      "endpoint": "/v1/YOUR_GATEWAY_ID/proxy/compat"
    },
    {
      "id": "x-ai/grok-beta",
      "name": "grok-beta",
      "endpoint": "/v1/YOUR_GATEWAY_ID/proxy/openrouter"
    }
   ]

   Important:

   • Replace YOUR_GATEWAY_ID with your actual Cloudflare AI Gateway ID
   • The id should match the provider/model identifier you configured in AI Gateway
   • The name is what users will use in API requests
   • The endpoint path should match your AI Gateway Universal Endpoint configuration

Usage

Standard Authentication (Bearer Token)

Use with any OpenAI-compatible client:

curl -H "Authorization: Bearer <PROXY_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-2.5-flash",
    "messages": [{"role": "user", "content": "Hello!"}]
  }' \
  https://your-worker.workers.dev/openai/chat/completions

List Available Models

curl -H "Authorization: Bearer <PROXY_API_KEY>" \
  https://your-worker.workers.dev/openai/models

Claude API Style

curl -H "Authorization: Bearer <PROXY_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-2.5-flash",
    "messages": [{"role": "user", "content": "Hello!"}],
    "max_tokens": 1024
  }' \
  https://your-worker.workers.dev/claude/v1/messages

JetBrains IDE Configuration

JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm, etc.) support Local AI integration. Since JetBrains doesn't support custom headers, use the URL-based authentication endpoint.

Setup Steps

1. Open Settings

   • Go to Settings → Tools → AI Assistant

2. Add Custom Model Provider

   • Click + to add a new provider
   • Select "OpenAI Compatible"

3. Configure Base URL

   https://your-worker.workers.dev/jb/<PROXY_API_KEY>
   

   Replace:

   • your-worker.workers.dev with your actual Worker URL
   • <PROXY_API_KEY> with your PROXY_API_KEY value

4. Model Selection

   • The models from your MODELS_CONFIG will be available
   • Select the model you want to use (e.g., gemini-2.5-flash)

5. Test Connection

   • Use the "Test Connection" button to verify
   • You should see a success message

Example Configuration

Base URL: https://my-ai-proxy.workers.dev/jb/my-secret-token

Notes

• No API key field needed (authentication is in the URL)
• All standard JetBrains AI features work (code completion, chat, etc.)
• Supports streaming responses

Project Structure

src/
├── index.ts              # Main entry point with auth and logging
├── constant.ts           # Shared constants and model definitions
├── utils.ts              # Utility functions
└── router/
    ├── OpenAIRouter.ts   # OpenAI-compatible endpoints
    └── ClaudeRouter.ts   # Claude-compatible endpoints

Environment Variables

Variable	Description	Required	Storage Type
CF_GATEWAY_KEY	Cloudflare AI Gateway API key	Yes	Secret (via wrangler secret or dashboard)
PROXY_API_KEY	Custom token for API authentication	Yes	Secret (via wrangler secret or dashboard)
MODELS_CONFIG	Supported models configuration	Yes	Plaintext (in wrangler.jsonc or .dev.vars)

License

MIT