Copilot Switch

Connect multiple OpenAI-compatible gateways and model providers to GitHub Copilot Chat.

English | 简体中文

Overview

Copilot Switch is a VS Code extension for users who want to use custom model providers inside GitHub Copilot Chat.

The original inspiration came from the need to connect Copilot Chat to an OpenAI-compatible API. This project focuses on a more flexible workflow: one extension can manage multiple gateway providers, multiple model entries, separate API keys, import/export, and model-level options from a visual Models Editor.

It is useful when you use several relay platforms, self-hosted gateways, local Ollama, or different API formats and want to switch between them from the Copilot Chat model picker.

Key Features

• Register custom models in the GitHub Copilot Chat model picker.
• Manage multiple gateway providers instead of being limited to one relay endpoint.
• Add, edit, enable, disable, import, export, and refresh model configurations from the Models Editor.
• Store API keys in VS Code SecretStorage instead of plaintext configuration files.
• Fetch model IDs from supported provider endpoints when configuring a gateway.
• Support OpenAI Chat Completions, OpenAI Responses, Ollama, Anthropic Messages, and Gemini API modes.
• Configure model-level context length, output tokens, reasoning options, headers, extra request body fields, vision support, and request delay.
• Use a selected model to generate Git commit messages from the Source Control view.

Screenshots

Models Editor

Add or Edit a Model

Enable Models in Copilot Chat

Requirements

• VS Code 1.111.0 or later.
• GitHub Copilot Chat installed and enabled.
• At least one provider API key or a local Ollama endpoint.

Quick Start

1. Install and enable Copilot Switch.
2. Run Copilot Switch: Open Models Editor from the command palette, or click the Copilot Switch status bar item.
3. Add a provider endpoint with baseUrl, apiMode, and API key.
4. Click Fetch Models when the provider supports model listing, or enter the model ID manually.
5. Save the model and enable it.
6. Open GitHub Copilot Chat, choose Manage Models..., select Copilot Switch, and enable the models you want to use.

Configuration Storage

Copilot Switch stores model configuration in the user profile:

~/.copilot/copilot-switch/models.json

API keys are stored separately in VS Code SecretStorage and referenced by apiKeyId. This keeps secrets out of the JSON configuration and makes it safer to import, export, or share model templates.

Minimal Configuration

Most users should use the Models Editor. Advanced users can edit the model JSON directly:

[
    {
        "displayName": "Qwen Coder via ModelScope",
        "id": "Qwen/Qwen3-Coder-480B-A35B-Instruct",
        "owned_by": "modelscope",
        "baseUrl": "https://api-inference.modelscope.cn/v1",
        "apiMode": "openai",
        "apiKeyId": "modelscope",
        "context_length": 256000,
        "max_tokens": 8192,
        "temperature": 0,
        "top_p": 1,
        "enabled": true
    }
]

API Modes

apiMode	Endpoint style	Typical use
openai	/chat/completions	OpenAI-compatible gateways such as ModelScope, SiliconFlow, DeepSeek gateways, and custom relay services
openai-responses	/responses	OpenAI Responses API and compatible gateways
ollama	/api/chat	Local or remote Ollama
anthropic	/v1/messages	Anthropic Claude API and compatible gateways
gemini	Gemini stream API	Google Gemini API and compatible gateways

Model Fields

Common fields:

• displayName: name shown in Copilot Chat and the Models Editor.
• id: provider model ID.
• owned_by: provider or gateway group ID.
• baseUrl: provider or gateway endpoint.
• apiMode: one of openai, openai-responses, ollama, anthropic, gemini.
• apiKeyId: SecretStorage key reference used to find the API key.
• configId: optional suffix for multiple configurations of the same model.
• family: optional Copilot model family hint.
• context_length, max_tokens, temperature, top_p: common generation parameters.
• vision: set to true for image-capable models.
• reasoning, reasoning_effort, thinking, enable_thinking, thinking_budget: reasoning options supported by different providers.
• headers: custom HTTP headers.
• extra: extra JSON fields merged into the request body.
• delay: model-specific delay in milliseconds.
• enabled: whether the model is available in the picker.
• useForCommitGeneration: whether this model can generate Git commit messages.

Development

npm install
npm run compile
npm run lint
npm run test
npm run build

Credits

Copilot Switch is inspired by and respectfully references oai-compatible-copilot. Thanks to the original project for exploring the OpenAI-compatible Copilot Chat provider idea.

Additional references:

• Hugging Face Chat Extension
• VS Code Chat Provider API

License

MIT License.