LM Proxy

An extension that enables external applications to access VSCode's GitHub Copilot capabilities through OpenAI and Anthropic compatible REST APIs, allowing you to leverage the power of GitHub Copilot outside of the VSCode environment. This extension utilizes the VSCode Language Model API (LM API) to communicate with the language models provided by GitHub Copilot.

Features

• External GitHub Copilot Access: Use GitHub Copilot's powerful AI capabilities from any application, not just within VSCode.
• OpenAI & Anthropic Compatible APIs: Access GitHub Copilot through industry-standard API formats that are compatible with OpenAI's and Anthropic's interfaces.
• Claude Code Support: Provides endpoints compatible with Claude Code for advanced AI assistance.
• Multiple Model Support: Seamlessly switch between different language models available in VSCode, including GitHub Copilot's models.
• Server Management: Easily start and stop the proxy server through the VSCode command palette or status bar.
• Streaming Support: Full support for streaming responses for real-time applications.
• Flexible Configuration: Customize the server port and log levels to fit your needs.

Installation

1. Open the Extensions view in VSCode.
2. Search for "LM Proxy".
3. Click Install.

Alternatively, you can download the .vsix file from the releases page and install it manually using the "Install from VSIX..." command.

Usage

Starting the Server

1. Open the Command Palette.
2. Run the LM Proxy: Start LM Proxy Server command.
3. The server status will be displayed in the status bar.

Stopping the Server

1. Open the Command Palette.
2. Run the LM Proxy: Stop LM Proxy Server command.

Selecting a Language Model

1. Open the Command Palette.
2. Run one of the following commands:
   • LM Proxy: Select OpenAI API Model
   • LM Proxy: Select Anthropic API Model
   • LM Proxy: Select Claude Code Background Model
   • LM Proxy: Select Claude Code Thinking Model
3. Choose your desired model from the list.

How Model Selection Works

• If you specify the model name as vscode-lm-proxy, the model selected in the extension settings will be used.
• If you specify a model name directly (e.g. gpt-4.1 or claude-3.5-sonnet), that model will be used for the request.

API Reference

The proxy server exposes the following endpoints:

OpenAI Compatible API

• Chat Completions: POST /openai/v1/chat/completions (supports streaming via the stream parameter)

curl -X POST http://localhost:4000/openai/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "vscode-lm-proxy",
    "messages": [{"role":"user","content":"Hello!"}],
    "stream": true
  }'

• List Models: GET /openai/v1/models

curl http://localhost:4000/openai/v1/models

• Retrieve Model: GET /openai/v1/models/{model}

curl http://localhost:4000/openai/v1/models/gpt-4.1

Anthropic Compatible API

• Messages: POST /anthropic/v1/messages (supports streaming via the stream parameter)

curl -X POST http://localhost:4000/anthropic/v1/messages \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "vscode-lm-proxy",
    "messages": [{"role":"user","content":"Hello!"}],
    "stream": true
  }'

• Count Tokens: POST /anthropic/v1/messages/count_tokens (counts the number of tokens in a message)

curl -X POST http://localhost:4000/anthropic/v1/messages/count_tokens \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "vscode-lm-proxy",
    "messages": [{"role":"user","content":"Hello, Claude"}]
  }'

• List Models: GET /anthropic/v1/models
• Retrieve Model: GET /anthropic/v1/models/{model}

Claude Code Compatible API

• Messages: POST /anthropic/claude/v1/messages (supports streaming via the stream parameter)

curl -X POST http://localhost:4000/anthropic/claude/v1/messages \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "vscode-lm-proxy",
    "messages": [{"role":"user","content":"Hello!"}],
    "stream": true
  }'

• Count Tokens: POST /anthropic/claude/v1/messages/count_tokens
• List Models: GET /anthropic/claude/v1/models
• Retrieve Model: GET /anthropic/claude/v1/models/{model}

For detailed information about the request and response formats, please refer to the official OpenAI API documentation and Anthropic API documentation.

Configuration

You can configure the extension settings in the VSCode settings UI or by editing your settings.json file.

• vscode-lm-proxy.port: The port number for the proxy server. (Default: 4000)
• vscode-lm-proxy.logLevel: The log level for the extension. (Default: 1 for INFO)
• vscode-lm-proxy.showOutputOnStartup: Whether to show the output panel on startup. (Default: false)

Commands

The following commands are available in the Command Palette:

• LM Proxy: Start LM Proxy Server: Starts the proxy server.
• LM Proxy: Stop LM Proxy Server: Stops the proxy server.
• LM Proxy: Select OpenAI API Model: Selects the OpenAI model to use.
• LM Proxy: Select Anthropic API Model: Selects the Anthropic model to use.
• LM Proxy: Select Claude Code Background Model: Selects the Claude Code background model to use.
• LM Proxy: Select Claude Code Thinking Model: Selects the Claude Code thinking model to use.
• LM Proxy: Show Output Panel: Shows the extension's output panel.
• LM Proxy: Clear Output Panel: Clears the extension's output panel.
• LM Proxy: Set Log Level: Sets the log level.

How It Works

LM Proxy leverages the VSCode Language Model API (LM API) to communicate with GitHub Copilot's language models. The extension acts as a bridge between external applications and VSCode's built-in language model capabilities:

1. The extension starts a local proxy server that implements OpenAI and Anthropic compatible API endpoints
2. When a request is received, it translates the request into the appropriate format for the VSCode Language Model API
3. The response from GitHub Copilot (via the LM API) is then converted back into the expected OpenAI or Anthropic format
4. This enables seamless integration with existing applications and tools that are designed to work with these popular API formats

This approach allows you to utilize the full power of GitHub Copilot in your applications without having to implement custom integrations.

License

This extension is licensed under the MIT License.