OAIProvider — OpenAI-Compatible Models in GitHub Copilot Chat

OAIProvider is a VS Code extension that connects any OpenAI-compatible API endpoint to GitHub Copilot Chat as a fully integrated language model provider — using the official VS Code LanguageModelChatProvider API.

Use it to bring models from NVIDIA NIM, Ollama, LM Studio, vLLM, Together AI, or any other OpenAI-format API into the Copilot Chat model picker.

✨ Features

• 🔌 Multiple providers — configure as many API endpoints as you need
• 🤖 Custom models — add any model IDs (no auto-discovery; you're in full control)
• 📡 Streaming — responses stream token-by-token via SSE, just like native Copilot models
• 🔑 Per-provider API keys — each provider has its own credentials
• 🛠️ Guided UI — step-by-step command palette wizards; no JSON editing required
• 🔄 Live updates — model list refreshes automatically when settings change

📋 Requirements

🚀 Quick Start

1 — Install the extension

From the VS Code Marketplace (recommended):

1. Open VS Code and go to the OAIProvider page, or search for OAIProvider in the Extensions panel (⇧⌘X)
2. Click Install
3. Reload VS Code (⇧⌘P → Developer: Reload Window)

From a VSIX file:

1. Download oai-provider-x.x.x.vsix from Releases
2. Run in your terminal:

code --install-extension oai-provider-x.x.x.vsix

3. Reload VS Code

From source:

git clone https://github.com/calganaygun/copilot-oai-provider.git
cd copilot-oai-provider
npm install
npm run compile
# Press F5 inside VS Code to launch Extension Development Host

2 — Add a provider

Open the Command Palette (⇧⌘P on macOS / Ctrl+Shift+P on Windows/Linux):

OAIProvider: Add Provider

Follow the 4-step wizard:

1. Display Name — e.g. NVIDIA NIM
2. Provider ID — slug, no spaces, e.g. nvidia-nim
3. Base URL — e.g. https://integrate.api.nvidia.com/v1
4. API Key — your bearer token (masked input; leave empty if not needed)

3 — Add a model

OAIProvider: Add Model

1. Pick the provider you just created
2. Enter the Model ID as the API expects (e.g. moonshotai/kimi-k2.5)
3. Enter a Display Name shown in the Copilot picker (e.g. Kimi K2.5)
4. Set max input tokens (e.g. 131072)
5. Choose whether the model supports tool calling

4 — Use in Copilot Chat

Open Copilot Chat → click the model picker → your model appears as:

Kimi K2.5 (NVIDIA NIM)

Send a message and enjoy streaming inference from your custom endpoint! 🎉

🔧 All Commands

Type OAIProvider in the Command Palette to see all commands:

⚙️ Configuration

All data is stored in openai-compat-provider.providers in VS Code's global settings. You can edit it directly in Settings JSON or via the guided commands above.

Example settings.json snippet:

"openai-compat-provider.providers": [
  {
    "id": "nvidia-nim",
    "displayName": "NVIDIA NIM",
    "baseUrl": "https://integrate.api.nvidia.com/v1",
    "apiKey": "nvapi-xxxxxxxxxxxx",
    "models": [
      {
        "id": "moonshotai/kimi-k2.5",
        "name": "Kimi K2.5",
        "maxInputTokens": 131072,
        "maxOutputTokens": 8192,
        "supportsToolCalling": true
      },
      {
        "id": "nvidia/llama-3.1-nemotron-ultra-253b-v1",
        "name": "Nemotron Ultra 253B",
        "maxInputTokens": 128000,
        "maxOutputTokens": 4096,
        "supportsToolCalling": true
      }
    ]
  },
  {
    "id": "ollama-local",
    "displayName": "Ollama (local)",
    "baseUrl": "http://localhost:11434/v1",
    "apiKey": "",
    "models": [
      {
        "id": "llama3.2",
        "name": "Llama 3.2 (local)",
        "maxInputTokens": 32000,
        "maxOutputTokens": 4096,
        "supportsToolCalling": false
      }
    ]
  }
]

🌐 Compatible Providers

🏗️ How It Works

This extension uses the VS Code LanguageModelChatProvider API to register itself as a first-class language model source. When Copilot Chat sends a message to one of the extension's models:

1. VS Code calls provideLanguageModelChatResponse with the conversation messages
2. The extension converts the VS Code message format to OpenAI's {"role", "content"} format
3. A fetch request is made to <baseUrl>/chat/completions with stream: true
4. The SSE response is parsed line-by-line and each chunk is reported back to VS Code via progress.report(new LanguageModelTextPart(...))
5. Copilot Chat renders the streaming response in real time

🛠️ Development

git clone https://github.com/calganaygun/copilot-oai-provider.git
cd copilot-oai-provider
npm install

Project Structure

src/
├── extension.ts   # Activation, provider registration, settings watcher
├── provider.ts    # LanguageModelChatProvider implementation (streaming, SSE)
├── commands.ts    # All 6 command handlers with step-by-step UI
├── config.ts      # Read/write helpers for VS Code settings
└── types.ts       # Shared TypeScript interfaces

🤝 Contributing

PRs and issues welcome! Please open an issue first for large changes.

📄 License

MIT — see LICENSE for details.