Addi is a VS Code extension that bridges GitHub Copilot with your preferred AI providers. Instead of being locked into a single vendor's models, you can connect Copilot to OpenAI-compatible APIs, Anthropic Claude, Google Gemini, DeepSeek, or any custom LLM endpoint — giving you full control over your AI stack while keeping Copilot's native interface and workflow.
💡 Why Addi? You get Copilot's seamless IDE integration (inline completions, agent mode, tool calling, file operations) with your choice of models — no vendor lock-in, no browser tabs, no CLI tools.
Connect OpenAI (GPT-4o, o-series), Anthropic (Claude Sonnet/Opus), Google (Gemini Pro/Flash), DeepSeek, Ollama (local), or any OpenAI-compatible endpoint. Each provider can host multiple models with independent configurations.
Models appear directly in Copilot's model picker — use them in chat, agent mode, or inline edits. Works with Copilot's built-in tools: code search, terminal execution, file editing, and diagnostics.
Real-time token-by-token output with reasoning/thinking display for supported models (OpenAI o-series, Anthropic Claude with extended thinking, Google Gemini thinking, DeepSeek R1). Supports reasoning_content backfill in multi-turn conversations — tested with DeepSeek V4/R1 and Xiaomi MiMo v2+ via OpenAI-compatible endpoint.
Send images to multimodal models and leverage Copilot's tool ecosystem. Models with tool-calling capabilities can use code search, terminal, file operations, and more directly from the chat interface.
Track real-time performance metrics per model — average speed, rolling window history, and latency stats. See exactly how your models perform in the sidebar.
React + TypeScript + Vite forms for provider and model editing, with per-provider conditional rendering (reasoning effort for OpenAI, thinking level for Anthropic/Google), field validation, and experimental features section.
Fully localized extension UI with English and Chinese (Simplified) support. Both code-level strings and contribution point strings (commands, views, configuration) are translated. Automatically matches VS Code's display language.
Backup, migrate, or share provider configurations as JSON. Supports encrypted export for secure configuration sharing across devices.
- Install the extension from the VS Code Marketplace
- Open the Addi sidebar — click the icon in the Activity Bar
- Add a Provider — click the
+button, fill in the API endpoint and key in the webview form - Add a Model — click a provider's
+button, configure model ID, capabilities, and reasoning options - Open Copilot Chat — select your model from the model picker and start coding
For detailed setup instructions, see the User Guide.
| Provider | Type ID | Examples |
|---|---|---|
| OpenAI-compatible | openai-completions |
OpenAI, DeepSeek, local Ollama, etc. |
| OpenAI Responses | openai-responses |
OpenAI (native tool support) |
| Anthropic | anthropic-messages |
Claude Sonnet, Claude Opus |
google-generateContent |
Gemini Pro, Gemini Flash |
| Document | Description |
|---|---|
| User Guide | Installation, configuration, usage, and troubleshooting |
| Changelog | Version history and release notes |
| Architecture Spec | Layered architecture, data flow, core design constraints |
| Coding Standards | Type safety, logging, error handling, naming conventions |
| AI SDK Reference | AI SDK v6 API types and mapping |
| VS Code API Reference | VS Code Copilot API and Proposed API reference |
| Reasoning Architecture | Architecture reference for reasoning/thinking capabilities |
| Config Export/Import | Encrypted JSON backup, migration, and sharing of configs |
- 🌐 Complete i18n: Full localization with Chinese (Simplified) support — everything from command labels to config descriptions
- 🖥️ Webview UI: New React + TypeScript + Vite forms replacing the legacy HTML editor
- 🧠 Reasoning Support: Thinking/reasoning content adaptation middleware and utilities
- 🔬 E2E Tests: Comprehensive test suite for data normalization and model fetching
- 🔐 Encrypted Config Export/Import: Secure backup and migration of provider/model configurations
- ⚡ Performance Metrics: Real-time speed monitoring and historical performance tracking
- 🛠️ More Providers: Ongoing integration of new providers and model types based on user demand
See the full changelog for details.
Issues and pull requests are welcome at github.com/deepwn/addi. Please review our coding standards before contributing.