Smoky Agent

Smoky Agent is an advanced, modern AI coding assistant directly integrated into VS Code. Powered by the incredibly capable @github/copilot-sdk underneath, Smoky gives your AI agent complete, secure access to your workspace and tools, turning it into a fully capable, autonomous coding partner!

Features

• Modern Webview Interface: A sleek, dark-themed, highly responsive chat UI supporting markdown, code syntax highlighting, word-by-word streaming, and a custom model selector.
• Native Tooling: Leveraging the Copilot SDK, Smoky natively understands terminal execution, file system operations, and workspace context out-of-the-box.
• Chrome DevTools Integration: Smoky comes with chrome-devtools-mcp built-in, allowing it to autonomously inspect, debug, and interact with the browser directly from your chat.
• Dynamic Configuration: Configure custom system messages, core rules, multiple LLM providers (including local models), and additional MCP servers effortlessly using your global config.json.
• Intelligent Abort: Safely cancel long-running AI generations natively from the interface with the push of a button.

Configuration (config.json)

Smoky Agent is fully configurable via a global config.json file securely stored in your user directory at ~/.smoky/config.json (e.g., C:\Users\username\.smoky\config.json on Windows). This ensures your settings, models, and history persist across VS Code workspaces and extension updates!

How to Edit the Config

1. Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P).
2. Type and run Smoky: Open Configuration (or access it from the settings icon in the webview title menu).
3. This will instantly open your global config.json file in the editor.
4. Note: Any changes you make to models or API keys will automatically reflect in the chat UI the moment you hit Save!

Configuration Schema

Here is an example config.json you can use to connect to local (e.g., Ollama/LMStudio) or remote models:

{
  "systemMessage": "You are a helpful and intelligent AI assistant.",
  "rules": [
    "Always provide concise answers.",
    "Use markdown for all your responses."
  ],
  "models": [
    {
      "name": "Local Ollama/LMStudio",
      "url": "http://127.0.0.1:11434/v1",
      "apiKey": "your-api-key-here",
      "model": "gpt-4"
    },
    {
      "name": "OpenAI GPT-4",
      "url": "https://api.openai.com/v1",
      "apiKey": "your-openai-api-key",
      "model": "gpt-4-turbo"
    }
  ],
  "mcpServers": {
    "my-custom-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
    }
  }
}

Configuration Options

• systemMessage (string): The core directive given to the agent.
• rules (string[]): An array of strict guidelines the agent must follow.
• models (array): A list of supported LLMs to use. You can easily switch between these models from the dropdown inside the Smoky Chat view.
  • name: Display name in the dropdown UI.
  • url: The base URL for the OpenAI-compatible API endpoint (e.g., local endpoints like Ollama/vLLM or external APIs).
  • apiKey: API Key for the model.
  • model: The exact model identifier.
• mcpServers (object): A map of additional Model Context Protocol (MCP) servers to load.
  • Note: Use type instead of transport (e.g., "type": "stdio"). Copilot SDK requires this formatting. All tools from these servers will automatically be granted to Smoky.

Usage

1. Open the Smoky Agent view from the Activity Bar (or run Smoky: Start).
2. Use the dropdown to select your configured model.
3. Chat with the agent! It can execute commands, edit files, and build apps for you safely and autonomously!