GitHub Copilot AI Gateway

Expose GitHub Copilot's AI models as a local HTTP server with OpenAI- and Anthropic-compatible endpoints.

Intended Use

This project is provided for local experimentation and learning on your own machine (or your own VM on the same host) and is not intended to be run for other users, teams, external clients, or as a hosted/shared service. It is not affiliated with or endorsed by Microsoft or GitHub. You are solely responsible for your use, including ensuring it complies with all applicable terms and policies (for example: GitHub Copilot terms, Visual Studio Code license and Marketplace terms, your organization’s policies, and any third‑party client terms). Model availability, entitlements, and usage limits still depend on your Copilot plan and organization settings. This software is provided “AS IS”, without warranties of any kind, and this documentation is not legal advice; if you are unsure whether your intended use is permitted, review the applicable terms and consult your organization or qualified counsel.

Why

Experiment with OpenAI-compatible / Anthropic-compatible clients against the models available through your existing GitHub Copilot in VS Code.

How It Works

graph LR
    A[Your App] -->|HTTP Request| B[Gateway Extension<br/>localhost:3000]
    B -->|vscode.lm API| C[GitHub Copilot]
    C -->|Stream Response| B
    B -->|HTTP Response| A
    
    style B fill:#0078d4,stroke:#fff,stroke-width:2px,color:#fff
    style C fill:#28a745,stroke:#fff,stroke-width:2px,color:#fff

The gateway runs inside VS Code and translates OpenAI- and Anthropic-compatible requests to VS Code's Language Model API.

⚠️ VS Code must be running with the extension enabled for the gateway to work.

Quick Start

1. Build & Install

./build.sh        # macOS/Linux
build.bat         # Windows

2. Open VS Code
Extension auto-starts. Check status bar (bottom-right): AI Gateway :3000 [model]

3. Get Your API Key
Click the status bar → "Copy API Key" → Paste into your environment:

export GHCGTW_API_KEY='your-api-key-here'

4. Try It

# Quick test with curl (replace YOUR-API-KEY)
curl http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR-API-KEY" \
  -d '{"model": "gpt-5-mini", "messages": [{"role": "user", "content": "Hi!"}]}'

# Or use the Python CLI
pip install -r requirements.txt
./qchat.py "Explain async/await in Python"

Settings

Setting	Default	Description
`ghcgtw.port`	`3000`	Port for the gateway server
`ghcgtw.defaultModel`	`gpt-5-mini`	Default model when client doesn't specify one
`ghcgtw.secondaryBindAddress`	(empty)	Additional IP to bind to (for VMs). See OpenClaw/Clawdbot Setup

Settings auto-apply — no restart required.

Security

Localhost by default: Gateway binds to 127.0.0.1 only
Secondary address: Optional binding to a private IP (for VMs). Public IPs are blocked.
API Key: Every request (except /health) requires Authorization: Bearer <KEY> or X-Api-Key: <KEY>
Browser protection: Blocks requests with Origin or Referer headers
Key storage: Auto-generated, stored encrypted in OS keychain

Note: The same API key is shared across all VS Code windows for the same user.

API Endpoints

Endpoint	Method	Format	Description
`/v1/messages`	POST	Anthropic	Chat completions (for Claude Code)
`/v1/chat/completions`	POST	OpenAI	Chat completions (for Crush, aider, etc.)
`/v1/models`	GET	OpenAI	List available models
`/v1/tokens`	POST	Custom	Count tokens for messages
`/health`	GET	Custom	Server status

Supported Features:

✅ Streaming responses (SSE)
✅ Multiple models (GPT-4o, Claude 3.5 Sonnet, o1, etc.)
✅ Token counting (/v1/tokens)

Not Supported (vscode.lm limitations):

❌ Function/tool calling (not wired to Copilot)
❌ Vision/image input (not exposed via API)
❌ Token usage in responses (only pre-request counting)
❌ Embeddings

Behavior

Authentication

All clients use the same API key (get it from the status bar popup).

Header	Format
`Authorization: Bearer <KEY>`	Standard (OpenAI SDK, Claude Code)
`X-Api-Key: <KEY>`	Alternative (some Anthropic clients)

Model Selection

Client specifies model AND it exists → Use that model
Client specifies model but NOT found → Use gateway default
Client doesn't specify model → Use gateway default

Claude Code and Crush will use whatever model you configure in the Gateway settings (e.g., claude-sonnet-4), regardless of what model name they send.

Client Configuration

Claude Code

Claude Code uses the Anthropic Messages API format (/v1/messages). We use "Foundry mode" to point Claude Code at a custom base URL (this local gateway) instead of the default Anthropic endpoint.

Note: Claude Code and other tools have their own terms and policies. Ensure your configuration and usage complies with those terms.

Option 1: Settings File (recommended)

Create or edit ~/.claude/settings.json:

{
  "env": {
    "CLAUDE_CODE_USE_FOUNDRY": "1",
    "ANTHROPIC_FOUNDRY_BASE_URL": "http://localhost:3000/anthropic/",
    "ANTHROPIC_FOUNDRY_API_KEY": "your-gateway-api-key",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4.5",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4.5",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-4.5"
  }
}

Option 2: Environment Variables

export CLAUDE_CODE_USE_FOUNDRY=1
export ANTHROPIC_FOUNDRY_BASE_URL=http://localhost:3000/anthropic/
export ANTHROPIC_FOUNDRY_API_KEY=your-gateway-api-key
export ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-4-5
export ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-4-5
export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-4-5

Verify it works:

claude "Say hello"

Note: CLAUDE_CODE_USE_FOUNDRY=1 routes Claude Code to a custom base URL. The gateway exposes /anthropic/v1/messages to match what Foundry mode expects. Requests are served via VS Code's vscode.lm API and the models available to your Copilot subscription.

Crush

Crush is a terminal-based AI coding assistant from Charm. It supports custom providers.

Create or edit ~/.config/crush/crush.json:

{
  "$schema": "https://charm.land/crush.json",
  "providers": {
    "copilot-gateway": {
      "name": "GHCGTW - Copilot Gateway",
      "type": "anthropic",
      "base_url": "http://localhost:3000",
      "api_key": "$GHCGTW_API_KEY",
      "models": [
        {
          "id": "claude-opus-4.5",
          "name": "GHCGTW - Claude Opus 4.5",
          "context_window": 200000,
          "default_max_tokens": 16000
        },
        {
          "id": "claude-sonnet-4.5",
          "name": "GHCGTW - Claude Sonnet 4.5",
          "context_window": 200000,
          "default_max_tokens": 16000
        },
        {
          "id": "gpt-5",
          "name": "GHCGTW - GPT 5",
          "context_window": 128000,
          "default_max_tokens": 16000
        },
        {
          "id": "gpt-5-mini",
          "name": "GHCGTW - GPT 5 Mini",
          "context_window": 128000,
          "default_max_tokens": 16000
        }
      ]
    }
  }
}

Note: The model ID must match an available Copilot model (e.g., claude-opus-4.5, gpt-5-mini). The gateway will return an error if the requested model doesn't exist. You can also use type: "openai-compat" with base_url: "http://localhost:3000/v1" if preferred.

Set your API key (get it from VS Code status bar → AI Gateway → Copy API Key):

# Temporary (current session only)
export GHCGTW_API_KEY=your-gateway-api-key

# Persistent (add to shell config)
echo 'export GHCGTW_API_KEY="your-gateway-api-key"' >> ~/.zshrc && source ~/.zshrc

Verify it works:

crush
# Press Ctrl+O to select "Copilot Gateway" as your provider

OpenClaw / Clawdbot

OpenClaw (formerly Moltbot/Clawdbot) is a personal AI assistant that runs on your own devices and connects to messaging platforms (WhatsApp, Telegram, Slack, Discord, etc.). It supports custom model providers, making it fully compatible with this gateway.

Prerequisites:

Node.js ≥22
OpenClaw installed: npm install -g openclaw@latest (or clawdbot@latest)

Configuration:

Create or edit ~/.openclaw/openclaw.json (or ~/.clawdbot/clawdbot.json):

{
  "models": {
    "mode": "replace",
    "providers": {
      "ghcgtw": {
        "baseUrl": "http://localhost:3000/v1",
        "apiKey": "YOUR_API_KEY_HERE",
        "api": "openai-completions",
        "authHeader": true,
        "headers": {
          "X-Api-Key": "YOUR_API_KEY_HERE"
        },
        "models": [
          {
            "id": "gpt-5-mini",
            "name": "GPT-5 Mini (via Gateway)",
            "reasoning": false,
            "input": ["text"],
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
            "contextWindow": 128000,
            "maxTokens": 16384
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "ghcgtw/gpt-5-mini"
      },
      "models": {
        "ghcgtw/gpt-5-mini": { "alias": "gpt5" }
      }
    }
  },
  "gateway": {
    "mode": "local"
  }
}

⚠️ Important: Replace YOUR_API_KEY_HERE in both places (the apiKey field AND inside headers.X-Api-Key). Get the key from VS Code status bar → AI Gateway → Copy API Key.

Note: Use "api": "openai-completions" (recommended) or "api": "anthropic-messages". The gateway supports both formats. The baseUrl must include /v1.

Test it:

openclaw doctor           # or: clawdbot doctor
openclaw agent --agent main --message "What model are you?"

⚠️ Note: You must specify --agent main (or -a main). Running just openclaw agent -m "..." will fail with a session selection error.

For VMs: If running OpenClaw/Clawdbot in a VM, see README.moltbot.md for detailed setup.

Other Clients

Any OpenAI-compatible or Anthropic-compatible client can use this gateway.

OpenAI SDK (Python):

import openai
import os

client = openai.OpenAI(
    base_url="http://localhost:3000/v1",
    api_key=os.environ.get('GHCGTW_API_KEY')
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
    stream=True
)

for chunk in response:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

aider:

export OPENAI_API_BASE=http://localhost:3000/v1
export OPENAI_API_KEY=$GHCGTW_API_KEY
aider --model gpt-4o

cURL:

# OpenAI format
curl http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $GHCGTW_API_KEY" \
  -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hi!"}]}'

# Anthropic format
curl http://localhost:3000/v1/messages \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: $GHCGTW_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{"model": "claude-sonnet-4", "max_tokens": 1024, "messages": [{"role": "user", "content": "Hi!"}]}'

Requirements

GitHub Copilot subscription
VS Code with Copilot extension (installed & authenticated)
VS Code must be running for gateway to work

For AI Agents: Integration Spec

Copy this section into your project's CLAUDE.md or AGENTS.md to enable AI integration.

Base URL: http://localhost:3000

Authentication: Authorization: Bearer <KEY> or X-Api-Key: <KEY> (from env GHCGTW_API_KEY)

Endpoints:

Endpoint	Format	Spec
`POST /v1/chat/completions`	OpenAI Chat Completions API	Streaming (SSE) and non-streaming
`POST /v1/messages`	Anthropic Messages API	Streaming (SSE) and non-streaming
`GET /v1/models`	OpenAI Models API	List available models
`GET /health`	—	No auth required

Behavior:

Model field is optional; gateway uses its configured default if omitted or not found
Streaming recommended for responsive UX

Not Supported:

Function/tool calling
Vision/image input
Token usage in responses

Compliance

This extension uses the official VS Code Language Model API (vscode.lm) as documented in Microsoft's Copilot Extensibility docs.

This repository does not grant you additional rights to use GitHub Copilot or any model.
Make sure your usage (including any third-party client you connect) complies with GitHub Copilot terms, VS Code license/Marketplace terms, and your organization’s policies.
Do not expose the gateway publicly or operate it as a shared/hosted service.

Troubleshooting

Issue	Fix
Connection refused	VS Code must be running. Check status bar shows `AI Gateway :3000 [model]`.
Unauthorized error	Set API key: `export GHCGTW_API_KEY='key'`. Get key from status bar popup.
Port in use	Only one VS Code instance can host. Non-hosting windows show `AI Gateway :3000` (without model).
No models	Sign in to GitHub Copilot extension

Authors

Matej Ciesko
Christopher Earley

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
.vscode		.vscode
ghcgtw		ghcgtw
.gitignore		.gitignore
AUTHORS		AUTHORS
LICENSE		LICENSE
README.md		README.md
README.moltbot.md		README.moltbot.md
TODO.md		TODO.md
build.bat		build.bat
build.sh		build.sh
qchat.py		qchat.py
requirements.txt		requirements.txt

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

GitHub Copilot AI Gateway

Intended Use

Why

How It Works

Quick Start

Settings

Security

API Endpoints

Behavior

Authentication

Model Selection

Client Configuration

Claude Code

Crush

OpenClaw / Clawdbot

Other Clients

Requirements

For AI Agents: Integration Spec

Compliance

Troubleshooting

Authors

About

Uh oh!

Releases

Packages

Contributors 2

Uh oh!

Languages

License

ciesko/ghcgtw

Folders and files

Latest commit

History

Repository files navigation

GitHub Copilot AI Gateway

Intended Use

Why

How It Works

Quick Start

Settings

Security

API Endpoints

Behavior

Authentication

Model Selection

Client Configuration

Claude Code

Crush

OpenClaw / Clawdbot

Other Clients

Requirements

For AI Agents: Integration Spec

Compliance

Troubleshooting

Authors

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Contributors 2

Uh oh!

Languages

Packages