README.md

---
license: mit
---
# gguf-code: IDE AI Coding Assistant

Monster coding agent lives in your IDE - moving beyond a CLI bridge to a fully native IDE AI assistant.

![screenshot](https://raw.githubusercontent.com/mochiyaki/code/master/demo.gif)

Shipped to vscode marketplace (currently unpublished; download it (.vsix) from [open-vsx](https://open-vsx.org/extension/gguf/gguf-code) and `code --install-extension gguf.gguf-code-x.x.x.vsix`) or it's compatible IDEs, i.e., cursor, windsurf, antigravity, etc., [marketplaces](https://open-vsx.org/extension/gguf/gguf-code)

## Features

- **Native Chat Interface**: Deeply integrated webview clean ui for smooth chat interactions inside VS Code (and/or its clone/fork IDEs).
- **Embedded AI Engine**: Run Coder's powerful AI logic directly within the extension process - no external CLI tool required.
- **Project-Aware Context**: Automatically understands your workspace structure and active files.
- **Smart Tools**:
    - `read_file`: Intelligently reads file contents with progressive loading for large files.
    - `write_file`, `find_files`, and more (Coming Soon).
- **Live Diff Preview**: Preview AI-suggested code changes with standard VS Code diff views.
- **Diagnostics Integration**: Uses LSP diagnostics to help fix errors in your code.

## Requirements

- Visual Studio Code version 1.104.0 or higher
- `coder.config.json` in your workspace root (for configuring AI providers).

## Getting Started

1.  **Installation**: Install the extension from the VS Code Marketplace or via `.vsix`.
2.  **Configuration**: Ensure you have a `coder.config.json` in your project root (see example below, can be any OpenAI compatible API or local/self-hosted API).

    lmstudio (localhost) example:
    ```json
    {
      "coder": {
        "providers": [
          {
            "name": "lmstudio",
            "models": [
              "openai/gpt-oss-20b"
            ],
            "baseUrl": "http://localhost:1234/v1"
          }
        ]
      }
    }
    ```
    
    ollama (localhost) example:
    ```json
    {
      "coder": {
        "providers": [
          {
            "name": "ollama",
            "models": [
              "gpt-oss:20b"
            ],
            "baseUrl": "http://localhost:11434/v1"
          }
        ]
      }
    }
    ```

    fetch from any api provider, see example below:
    ```json
    {
      "coder": {
          "providers": [
            {
              "name": "openrouter",
              "baseUrl": "https://openrouter.ai/api/v1",
              "apiKey": "your-api-key",
              "models": ["anthropic/claude-4.5-sonnet"]
            }
          ]
      }
    }
    ```
    
3.  **Launch**: Click the "Code" monster icon in the Activity Bar (left sidebar) to open the chat.

## Usage

- **Chat**: Type naturally in the chat window. "Explain this file", "Fix this bug", etc.
- **Context**: The assistant automatically knows about your open files.

## Architecture

This extension has migrated from a thin client (CLI bridge) to a thick client architecture:

- **Frontend**: React + Vite (Webview).
- **Backend**: Node.js Extension Host (running `ai-sdk-client`).
- **Communication**: VS Code Message Passing API.

## Development

```bash
npm install

# Build both Webview and Extension
npm run build

# Watch mode (rebuilds on change)

# Terminal 1:
npm run watch:webview

# Terminal 2:
npm run watch
```

## Contributing

Contributions are welcome!

## License

MIT