Why?

• You have a lot of free quota on GitHub Copilot, you want to use it like OpenAI-compatible APIs.
• You want the computing power of GitHub Copilot beyond VS Code.
• You want to use modern models like gpt-4.1 free.
• You have multiple GitHub accounts and the free quota is just wasted.
• Host LLM locally and leave the computing remotely.

Features

• Proxies requests to https://api.githubcopilot.com
  • Support endpoints: /chat/completions, /models
• User-friendly admin UI:
  • Log in with GitHub and generate tokens
  • Add tokens manually
  • Manage multiple tokens with ease
  • View chat message usage statistics
  • Simple chat bot for model evaluation
    • Client-side chat session history

How to use

• Start the proxy server
  • Option 1: Use Docker

    docker run -p 3000:3000 ghcr.io/coxy-proxy/coxy:latest

  • Option 2: Use pnpx(recommended) or npx

    pnpx coxy

• Browse http://localhost:3000 to generate the token by following the instructions.
  • Or add your own token manually.
• Set a default token.
• Your OpenAI-compatible API base URL is http://localhost:3000/api
  • You can test it like this: (no need authorization header since you've set a default token!)

  curl --request POST --url http://localhost:3000/api/chat/completions --header 'content-type: application/json' \
  --data '{
      "model": "gpt-4",
      "messages": [{"role": "user", "content": "Hi"}]
  }'

  • You still can set a token in the request header authorization: Bearer <token> and it will override the default token.
• (Optional) Use environment variable PORT for setting different port other than 3000.

Available environment variables

• PORT: Port number to listen on (default: 3000)
• LOG_LEVEL: Log level of pino (default: info)
• DATABASE_URL: Database URL for Prisma (currently only supports sqlite). Should start with file:. (default: file:../coxy.db)
  • The relative path will be resolved to the absolute path at runtime.

Advanced usage

• Dummy token _ to make coxy use the default token.
  • In most cases, the default token just works without 'Authorization' header. But if your LLM client requires a non-empty API key, you can use the special dummy token _ to make coxy use the default token.
• Provisioning: launch the CLI with --provision to force initialize the database schema via Prisma.
• Tips for using docker:
  • Mount the sqlite db file from host to persist the tokens and use .env file to set environment variables. Use --provision first time to initialize the database schema via Prisma, e.g.

    docker run -p 3000:3000 -v /path/to/sqlite.db:/app/coxy.db -v /path/to/.env:/app/.env ghcr.io/coxy-proxy/coxy:latest --provision

Troubleshooting

• I got status code 400 when using GPT-5-mini model.
  • Make sure you have enabled "OpenAI GPT-5 mini" in your github account Copilot settings. The URL looks like https://github.com/settings/copilot/features.
• I got the error Fail to load API keys when opening API Keys page.
  • Make sure you have sqlite db file and provisioned correctly.
• I'm using podman and cannot access via localhost:3000 but can access via 127.0.0.1:3000.
  • It seems podman's known issue or bug. Just use 127.0.0.1:3000 instead or setting another host name, like 127.0.0.1 coxy, in your /etc/hosts file.

Use cases

• Use with LLM CLI locally.
• Chat with GitHub Copilot by Open WebUI.

Requirements

• Node.js 22 or higher

References

• https://www.npmjs.com/package/@github/copilot-language-server
• https://github.com/B00TK1D/copilot-api
• https://github.com/ericc-ch/copilot-api
• https://hub.docker.com/r/mouxan/copilot

Licensed under the MIT License.