Copilot Adapter

Extend VS Code's native Copilot Chat with third-party AI models. Switch between DeepSeek, MiniMax, and more directly from the Copilot model selector — the same experience as built-in Copilot models, with your own API keys, no Copilot subscription, and no local proxy needed.

简体中文

• Providers
• Quick Start
• Security
• Features
  • Thinking Modes
  • Vision Proxy
  • Per-Provider Settings
• Configuration Reference
• Commands

Providers

Quick Start

1. Install the extension.
2. Open the Language Models panel (Ctrl/Cmd+Shift+P → Language Models), select a provider, and enter your API key — this is the native VS Code approach and the recommended way.
3. Open Copilot Chat, click the model selector, and choose a model.

Alternative: Use the command Copilot Adapter: Add API Key (Ctrl/Cmd+Shift+P) to store a key without opening the Language Models panel.

A step-by-step walkthrough covers all of the above: open it via Ctrl/Cmd+Shift+P → Welcome: Open Walkthrough → AI Models Adapter for Copilot Chat.

For a visual guide with screenshots, see How to Add a Model Provider.

Security

API keys are stored exclusively in VS Code's Secret Storage, backed by the OS credential manager — Keychain on macOS, Credential Manager on Windows, libsecret on Linux.

• Never written to settings.json — keys are stored as secrets, completely separate from VS Code settings. They cannot be synced via Settings Sync and will never appear in any configuration file on disk.
• Cannot be accidentally committed — because keys exist only in the OS credential store, there is no file to stage or push.
• Masked in all logs — even when debugMode is set to verbose, API keys are masked in the output channel (e.g. sk-12345678••••••••cdef). The raw key value is never logged.

Features

Thinking Modes

Reasoning models (DeepSeek V4 Pro, MiniMax M-series) expose a configurable effort level, accessible in the model's settings inside the Language Models panel:

Levels shown are DeepSeek V4 as an example; option names may differ across providers.

Vision Proxy

Text-only models cannot accept image attachments directly. When a vision proxy is configured, the extension automatically describes any attached images using a separate vision-capable model and injects those descriptions as text — so text-only models can handle image attachments seamlessly.

Set up via Copilot Adapter: Set Vision Proxy Model or the copilot-adapter.visionProxyModel setting.
Disable at any time by setting the value to off.

Per-Provider Settings

Advanced customization — no configuration needed for normal use.

Each provider can be configured independently under copilot-adapter.providers:

"copilot-adapter.providers": {
  "deepseek": {
    "enabled": true,
    "temperature": 0.7
  },
  "minimax": {
    "enabled": false
  }
}

Configuration Reference

Debug Mode Levels

Commands

License

MIT