Xcode Proxy (xcode-proxy)

English | 中文

Xcode Proxy is a lightweight, high-performance AI proxy server designed specifically for Xcode. It acts as a "Universal Adapter" between Xcode and modern AI models.

Goal: To Unlock and Connect the closed Xcode ecosystem with powerful, diverse AI services like OpenRouter, DeepSeek, Anthropic, and more.

---

🏗 Design Philosophy & Architecture

Why Xcode Proxy?

Xcode's AI features (like Copilot) are often locked to specific providers. Developers are left stranded when they want to use cheaper, smarter, or more compliant models.

Our Design Philosophy:

1. Invisible: To Xcode, it looks exactly like a standard OpenAI/Copilot service. No hacking of Xcode binaries required.
2. Universal: Connects to any service that complies with the OpenAI API standard.
3. Dynamic: Configuration changes take effect immediately. Simply edit settings.toml to switch model routes without restarting or recompiling.
4. Resilient: Handles authentication, error forwarding, and provides intelligent fallback defaults automatically.

Architecture

graph TD
    User[Xcode / Cursor / VSCode] -->|Request: model='GLM-4'| Proxy[⚡️ Xcode Proxy :3000]
    
    subgraph "Xcode Proxy Core"
        Router[Smart Router]
        Config[Dynamic Config Settings.toml]
        Adapter[Protocol Adapter]
    end
    
    Proxy --> Router
    Config -.->|Load Rules| Router
    Router -->|Match Config| Adapter
    

    Adapter -->|Forward| Cloud1["☁️ OpenRouter"]
    Adapter -->|Forward| Cloud2["☁️ DeepSeek"]
    Adapter -->|Forward| Cloud3["☁️ Zhipu AI (GLM)"]
    
    Cloud1 -->|Response| Adapter
    Adapter -->|Stream| Proxy
    Proxy -->|Stream| User

---

🚀 Quick Start

1. Installation (Recommended)

Install as a Python package to use the xcode-proxy command globally.

# Run in project root
pip install .

# Or for development (Editable mode)
pip install -e .

2. Start Server

After installation, run:

xcode-proxy

Custom port:

xcode-proxy --port 8080

Or run via script directly:

python app/cli.py

---

⚙️ Configuration (settings.toml)

The project uses settings.toml for configuration (What You See Is What You Get).

[server]
host = "0.0.0.0"   # Listen address
port = 3000        # Server port

[server.models]
default = "GLM-4.7" # Default model if none specified by client

# --- Custom Models ---

# 1. Zhipu AI
[models.GLM-4]
type = "zhipu"
api_key = "your-api-key"

# 2. OpenRouter (Access global models)
[models."openrouter/free"]
type = "openai"
api_key = "sk-or-..."
api_base = "https://openrouter.ai/api/v1"

# 3. DeepSeek
[models.deepseek-chat]
type = "openai"
api_key = "sk-..."
api_base = "https://api.deepseek.com"

Smart Fallback: If ~/.claude/settings.json exists, the server automatically loads ANTHROPIC_API_KEY from it as a fallback model.

---

🛠️ Xcode Integration

Example (Copilot for Xcode):

1. Click "Add Provider" in the Host App.
2. Select "Local".
3. Enter the server port (default 3000) in the Port field.
4. Click Save.

That's it! You can now access all models configured in settings.toml directly within Xcode.

---

📝 License

MIT License