copilot2api

OpenAI-compatible API service backed by the GitHub Copilot SDK (@github/copilot-sdk).

Expose Copilot models through a drop-in OpenAI API, so any OpenAI-compatible client can talk to GitHub Copilot.

---

Requirements

• Node.js ≥ 20
• GitHub Copilot CLI installed and available in PATH (copilot)
  See installation guide
• A GitHub Copilot subscription (or BYOK configuration)

Installation

npm install
npm run build

Configuration

Set one of the following environment variables to authenticate with GitHub:

Variable	Description
COPILOT_GITHUB_TOKEN	GitHub token for Copilot (highest priority)
GITHUB_TOKEN	GitHub personal access token
GH_TOKEN	Alternate token variable
PORT	HTTP port (default: 3000)

If none of these are set the SDK falls back to the stored credentials from gh auth login or copilot auth login.

Image Compression

When images are included in chat messages, they are downloaded to a temporary local file before being sent to the Copilot session. The following options control whether those images are compressed first:

Variable	Description
IMAGE_COMPRESS_ENABLED	Enable image compression before upload (default: true). Set to false or 0 to disable.
IMAGE_COMPRESS_MAX_SIZE	Maximum width or height in pixels after resizing (default: 1920). The image is resized proportionally so neither dimension exceeds this value.

Image compression uses sharp, which works on Windows, Linux, and macOS.

Running the Server

# Development (tsx, no build step)
npm run dev

# Production (after `npm run build`)
npm start

API Endpoints

GET /v1/models

Returns the list of available Copilot models in OpenAI format.

curl http://localhost:3000/v1/models

POST /v1/chat/completions

Sends a chat completion request. Supports both streaming and non-streaming responses.

Non-streaming:

curl http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4.1",
    "messages": [
      {"role": "user", "content": "What is 2+2?"}
    ]
  }'

Streaming:

curl http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4.1",
    "messages": [
      {"role": "user", "content": "What is 2+2?"}
    ],
    "stream": true
  }'

JSON 模式 (格式化输出):

curl http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4-o",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant. Respond in valid JSON."},
      {"role": "user", "content": "Return a JSON object with the keys \"name\" and \"age\" for a fictional character."}
    ],
    "response_format": { "type": "json_object" }
  }'

图片分析与格式化输出:

curl http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "请分析这张图片的内容，并以 JSON 格式返回。要求包含两个字段: \"image_url\" (图片的 URL) 和 \"description\" (图片的详细描述)。"
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://raw.githubusercontent.com/github/copilot-sdk/main/docs/assets/copilot-sdk-logo.png"
            }
          }
        ]
      }
    ],
    "response_format": { "type": "json_object" }
  }'

多模态 (图片分析):

curl http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4-o",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "这幅图中有什么？" },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://wallpapers.com/images/hd/cute-cats-pictures-ofp9qyt72qck6jqg.jpg"
            }
          }
        ]
      }
    ]
  }'

Using with OpenAI SDK

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "http://localhost:3000/v1",
  apiKey: "not-needed", // any non-empty string works
});

const response = await client.chat.completions.create({
  model: "gpt-4.1",
  messages: [{ role: "user", content: "Hello!" }],
});

console.log(response.choices[0].message.content);

Architecture

OpenAI-compatible client
         ↓ HTTP (OpenAI API format)
   copilot2api (Express)
         ↓ @github/copilot-sdk (JSON-RPC)
   Copilot CLI (server mode)
         ↓
   GitHub Copilot backend

Each POST /v1/chat/completions request creates a fresh Copilot session. The full conversation history from the messages array is injected via the session's system message so the model always has the proper context.

License

MIT