English | 中文

Copilot Proxy

Convert GitHub Copilot into an OpenAI-compatible API with one click. Works with Continue and other plugins.

✨ Features

• One-Click Auth - Automatic GitHub OAuth device-flow on first run, no manual token copying
• Auto Refresh - Background automatic Copilot Token refresh, zero intervention needed
• Remember Login - Token saved locally, auto-restored on next launch
• OpenAI Compatible - Standard OpenAI API format, works with any compatible client
• Bypass Education/Free plan restrictions on models like Claude

📋 Prerequisites

1. GitHub account with one of the following enabled:

   • Copilot Free
   • Copilot Pro
   • Copilot Education

2. Python 3.10+

⬇️ Download & Install

Download releases

You can download prebuilt executables from the GitHub Releases page and run them directly:

• Windows: Copilot_Proxy-windows.exe

• macOS: Copilot_Proxy-macos

  Make executable and run:

  chmod +x Copilot_Proxy-macos
  ./Copilot_Proxy-macos

All release files are automatically built by GitHub Actions, source code is fully open and auditable

Build from source

1. Install dependencies

git clone https://github.com/Open-Copilot-Proxy/Copilot_Proxy.git
cd Copilot_Proxy
pip install -r requirements.txt

2. Run the Script

python main.py

🚀 Quick Start

1. First-Time Authorization

The script will automatically open your browser. Follow the prompts:

[1/3] Requesting device verification code...

[2/3] Please complete authorization in your browser:

  ┌─────────────────────────────┐
  │                             │
  │     Code:  ABCD-1234        │
  │                             │
  └─────────────────────────────┘

  Open: https://github.com/login/device
  (Browser opened automatically)

[3/3] Waiting for authorization...

[✓] Authorization successful!

2. Configure Continue

Open Continue's config.yaml and add model configurations:

models:
    # Claude Sonnet
  - name: Claude Sonnet 4.6 (Copilot_proxy)
    provider: openai
    model: claude-sonnet-4.6
    apiBase: http://localhost:{PROXY_PORT}
    apiKey: "dummy"
    roles:
      - chat
      - edit

  # Claude Opus
  - name: Claude Opus 4.6 (Copilot_proxy)
    provider: openai
    model: claude-opus-4.6
    apiBase: http://localhost:{PROXY_PORT}
    apiKey: "dummy"
    roles:
      - chat

  # GPT-5.4
  - name: GPT-5.4 (Copilot_proxy)
    provider: openai
    model: gpt-5.4
    apiBase: http://localhost:{PROXY_PORT}
    apiKey: "dummy"
    roles:
      - chat

  # Gemini 3.1 Pro preview
  - name: Gemini 3.1 Pro Preview (Copilot_proxy)
    provider: openai
    model: gemini-3.1-pro-preview
    apiBase: http://localhost:{PROXY_PORT}
    apiKey: "dummy"
    roles:
      - chat

  # Code Completion (Tab)
  - name: Codex Mini (Copilot)
    provider: openai
    model: gpt-5.1-codex-mini
    apiBase: http://localhost:{PROXY_PORT}
    apiKey: "dummy"
    roles:
      - autocomplete

3. Start Using

Open the Continue sidebar in VS Code, select a model, and start chatting.

🔧 Advanced Usage

Change Port

Edit the top of config.py:

PROXY_PORT = 15432  # Change to your desired port

Re-authorize

Delete the token file and re-run:

rm .copilot_token.json
python main.py

Use as a Backend for Other Tools

Any tool that supports the OpenAI API format can connect:

API Base: http://localhost:15432
API Key:  any value (e.g. "dummy")
Model:    claude-sonnet-4.6

Model Fallback

• What it does: If a requested model is unavailable for your Copilot account (the Copilot API returns a model_not_supported error), the proxy can automatically retry the request using a supported fallback model so clients don't fail immediately.

• How it works: On startup the proxy selects a fallback model by querying the Copilot models endpoint (selection logic lives in fallback.py). The default priority list is defined in fallback.py (for example: gpt-4.1, gpt-4o, gpt-5-mini, raptor-mini). When the upstream returns a 400 with code model_not_supported, the proxy replaces the top-level model field in the client JSON with the chosen fallback model and retries once. The proxy also exposes the currently selected fallback via GET /fallback.

• Customize: Edit fallback.py to change the PREFERRED_PREFIXES or the selection logic. The script will pick the first usable model matching the configured priorities.

• Check current fallback:

curl -sS http://localhost:15432/fallback

• Testing: point a client at http://localhost:15432 and request an unsupported model (e.g., claude-mythos-preview) — watch the proxy logs for fallback messages.

🏗️ How It Works

Continue / Other Clients
        │
        │ POST /chat/completions (OpenAI format)
        ▼
┌───────────────────────┐
│  copilot_proxy.py     │  localhost:15432
│                       │
│  • OAuth device flow  │
│  • Auto token refresh │
│  • Request forwarding │
└───────────────────────┘
        │
        │ Bearer <copilot_token>
        ▼
GitHub Copilot API
        │
        ▼
Claude / GPT / Gemini

1. OAuth Authorization - Obtain a long-lived ghu_ token via GitHub device flow
2. Token Exchange - Exchange the ghu_ token for a short-lived Copilot token (~30 min validity)
3. Auto Refresh - Background refresh of the Copilot token every 25 minutes
4. Request Forwarding - Forward OpenAI-format requests to the Copilot API with valid credentials

❓ FAQ

Q: Copilot Token refresh failed?

Make sure your GitHub account has Copilot enabled: https://github.com/settings/copilot

Q: Model returns 401/403?

The token may have expired. Restart the script to auto-refresh.

Q: Some models are unavailable?

Different Copilot subscription tiers have access to different models. The free plan may not support all models.

Q: Can I use it on another computer?

Each computer needs to run the script and authorize separately. Do not copy .copilot_token.json to other machines.

Q: Is it secure?

• Tokens are only stored locally in your .copilot_token.json file
• All requests communicate directly with the official GitHub API
• No third-party servers involved

📄 License

MIT