Copilot Proxy

Overview

Copilot Proxy is a Visual Studio Code extension that exposes the VS Code Language Model API through an OpenAI-compatible Express server. This project is intended for research, experimentation, and local prototyping rather than production use.

This project is not affiliated with GitHub, Microsoft, or OpenAI.

Current GitHub Copilot model availability depends on your subscription and the VS Code Language Model API.

Acknowledgements

This repository is a rebranded and maintained fork of the original copilot-proxy project by Lutz Leonhardt.

Special thanks to Lutz Leonhardt for creating the original project and publishing it under the MIT License, which made this fork, rebranding, and continued maintenance possible.

Features

• Start and stop the local proxy server from inside VS Code.
• Configure the listening port through settings or a command.
• Forward chat requests with streaming and non-streaming support.
• Protect the proxy with locally stored API tokens.
• Expose OpenAI-compatible endpoints for models, chat completions, responses, and related assistant routes.

Installation

1. Download the latest .vsix package from the GitHub Releases page.
2. Open Visual Studio Code.
3. Open the Extensions view (Cmd+Shift+X on macOS or Ctrl+Shift+X on Windows/Linux).
4. Open the extensions menu (...) and choose Install from VSIX....
5. Select the downloaded .vsix file.
6. Reload VS Code if prompted.

Configuration

The extension exposes one primary setting:

• copilotProxy.port — default: 3000

You can change it either:

• in the VS Code Settings UI by searching for Copilot Proxy, or
• from the Command Palette with Copilot Proxy: Configure Port.

API token authentication

The extension includes token management commands so you can lock down your local proxy.

Manage tokens

1. Create a token with Copilot Proxy: Create API Token.
2. List saved tokens with Copilot Proxy: List API Tokens.
3. Remove a token with Copilot Proxy: Remove API Token.

Tokens are stored in VS Code global state and reused across sessions.

Use tokens in requests

When at least one token exists, requests must include a bearer token in the Authorization header:

curl -H "Authorization: Bearer cpx_your_token_here" \
     -H "Content-Type: application/json" \
     -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}]}' \
     http://localhost:3000/v1/chat/completions

If no tokens have been created, authentication is disabled and the server accepts requests without token validation.

Using the extension

Start the server

• Run Copilot Proxy: Start Server from the Command Palette.
• The server starts on the configured port, which defaults to 3000.

Stop the server

• Run Copilot Proxy: Stop Server from the Command Palette.

Permission prompt

On the first model request, VS Code may ask you to grant permission for this extension to access the Language Model API. Accept the prompt so requests can complete successfully.

Models and embeddings

The proxy exposes OpenAI-style model discovery at /v1/models.

List models

• GET /v1/models returns all chat models plus any embedding models exposed by VS Code.
• GET /v1/models?type=chat returns only chat-capable models.
• GET /v1/models?type=embedding returns only embedding models.

Example requests:

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

curl http://localhost:3000/v1/models?type=chat

curl http://localhost:3000/v1/models?type=embedding

Embeddings endpoint

The proxy also exposes POST /v1/embeddings and forwards requests to the VS Code embeddings API:

curl -X POST http://localhost:3000/v1/embeddings \
  -H "Authorization: Bearer cpx_your_token_here" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "text-embedding-3-small",
    "input": "hello world"
  }'

If model is omitted, the proxy uses the first embedding model reported by VS Code.

Proposed API caveat

Embedding support depends on the VS Code Embeddings proposed API. That means chat completions can work while embedding discovery and /v1/embeddings do not.

This repository already declares the proposal in package.json via enabledApiProposals, but the runtime extension you have loaded must also include that manifest entry.

If GET /v1/models?type=embedding returns an empty list, or POST /v1/embeddings returns a 501 error, check the following:

1. You are running the latest build of this extension, not an older installed VSIX.
2. VS Code has been reloaded after installing or rebuilding the extension.
3. Your VS Code build supports the embeddings proposal.
4. If your setup requires it, start VS Code with:

code --enable-proposed-api hyorman.copilot-proxy

When the proposal is unavailable, the proxy degrades gracefully:

• GET /v1/models still lists chat models.
• GET /v1/models?type=embedding returns no embedding entries.
• POST /v1/embeddings returns a clear error instead of a generic server failure.

License

This repository is distributed under the MIT License. The upstream copyright notice is retained as required by the license.

Skills API

The extension exposes an OpenAI-compatible Skills API at /v1/skills. Skills are versioned file bundles anchored by a SKILL.md manifest.

Skill bundle format

Every bundle must include exactly one SKILL.md file with YAML frontmatter:

---
name: My Skill
description: One-line summary of the skill
---

Detailed instructions for the agent go here.

Additional files (code, data, configs) can be included alongside the manifest.

Endpoints

Method	Path	Description
POST	/v1/skills	Create a skill from multipart or zip upload
GET	/v1/skills	List skills (paginated)
GET	/v1/skills/:id	Retrieve a skill
POST	/v1/skills/:id	Update default version or metadata
DELETE	/v1/skills/:id	Delete a skill and its files
POST	/v1/skills/:id/versions	Upload a new version

Create a skill

curl -X POST http://localhost:3000/v1/skills \
  -H "Authorization: Bearer cpx_your_token" \
  -F "files[]=@SKILL.md" \
  -F "files[]=@helper.py"

Or upload a zip archive:

curl -X POST http://localhost:3000/v1/skills \
  -H "Authorization: Bearer cpx_your_token" \
  -F "files=@my-skill.zip"

Upload a new version

curl -X POST http://localhost:3000/v1/skills/skill_abc123/versions \
  -H "Authorization: Bearer cpx_your_token" \
  -F "files[]=@SKILL.md" \
  -F "files[]=@updated_helper.py"

Update default version

curl -X POST http://localhost:3000/v1/skills/skill_abc123 \
  -H "Authorization: Bearer cpx_your_token" \
  -H "Content-Type: application/json" \
  -d '{"default_version": 2}'

Using skills with the Responses API

Attach skills to a /v1/responses request via the skills field. You can use registered skill references or inline skills.

Skill reference (uses a skill registered via /v1/skills):

curl -X POST http://localhost:3000/v1/responses \
  -H "Authorization: Bearer cpx_your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "input": "Write a Python function to sort a list.",
    "skills": [
      { "type": "skill_reference", "skill_id": "skill_abc123" }
    ]
  }'

Specific version:

{ "type": "skill_reference", "skill_id": "skill_abc123", "version": 2 }

Inline skill (base64-encoded content, no registration required):

curl -X POST http://localhost:3000/v1/responses \
  -H "Authorization: Bearer cpx_your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "input": "Refactor this code.",
    "skills": [
      {
        "type": "inline",
        "name": "Code Style",
        "description": "Enforce coding standards",
        "source": {
          "type": "base64",
          "media_type": "text/markdown",
          "data": "QWx3YXlzIHVzZSBUeXBlU2NyaXB0IHN0cmljdCBtb2Rl"
        }
      }
    ]
  }'

Using skills with the Assistants API

Skills can be attached to assistants and runs. When a run executes, skill instructions are resolved and injected into the system prompt alongside the assistant's instructions.

Attach skills to an assistant

curl -X POST http://localhost:3000/v1/assistants \
  -H "Authorization: Bearer cpx_your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "name": "My Coding Assistant",
    "instructions": "You are a helpful coding assistant.",
    "skills": [
      { "type": "skill_reference", "skill_id": "skill_abc123" }
    ]
  }'

Override skills on a run

Skills specified on a run override the assistant's skills for that run:

curl -X POST http://localhost:3000/v1/threads/thread_abc123/runs \
  -H "Authorization: Bearer cpx_your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "assistant_id": "asst_abc123",
    "skills": [
      { "type": "skill_reference", "skill_id": "skill_xyz789", "version": "latest" }
    ]
  }'

If no skills are specified on the run, the assistant's skills are used. If the run specifies an empty skills array, no skills are injected.

Skills in thread-and-run creation

curl -X POST http://localhost:3000/v1/threads/runs \
  -H "Authorization: Bearer cpx_your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "assistant_id": "asst_abc123",
    "thread": {
      "messages": [{ "role": "user", "content": "Help me write tests." }]
    },
    "skills": [
      { "type": "skill_reference", "skill_id": "skill_abc123" },
      {
        "type": "inline",
        "name": "Test Style",
        "description": "Testing conventions",
        "source": {
          "type": "base64",
          "media_type": "text/markdown",
          "data": "VXNlIGRlc2NyaWJlL2l0IGJsb2NrcyB3aXRoIHZpdGVzdA=="
        }
      }
    ]
  }'

Update assistant skills

curl -X POST http://localhost:3000/v1/assistants/asst_abc123 \
  -H "Authorization: Bearer cpx_your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "skills": [
      { "type": "skill_reference", "skill_id": "skill_new456" }
    ]
  }'