megg

Git-native, contextual memory infrastructure that lives directly in your codebase.

megg turns stateless AI agents into long-term employees. It provides a standardized way for agents to read, store, and manage context directly within your project's folder structure.

Every session is fresh, but with megg, your agent remembers.

⚡️ Quick Start

1. Install

git clone https://github.com/opencode-experiment/megg.git
cd megg
npm install && npm run build

2. Configure (Claude Desktop / MCP Client)

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "megg": {
      "command": "node",
      "args": ["/absolute/path/to/megg/build/index.js"]
    }
  }
}

3. Instruct Your Agent

Add this to your system prompt:

Memory/Context: You have access to megg, a memory tools suite.

• At session start, call megg.awake() to orient yourself.
• When entering a new directory, call megg.recall(path).
• If you make a significant decision, call megg.remember(path, content).

📚 Documentation

• User Manual: How to structure memory, best practices, and the "Good Employee" philosophy.
• API Reference: Detailed documentation of the MCP tools (awake, recall, settle, etc.).
• Design: The core philosophy and architecture.

🏗️ How It Works

megg inserts .megg/ directories at key boundaries in your project (e.g., root, src/, api/).

project/
├── .megg/              # Root memory
│   ├── info.md         # Identity & Rules
│   └── map.md          # Navigation (auto-generated)
├── src/
│   ├── api/
│   │   ├── .megg/      # API memory
│   │   │   ├── info.md
│   │   │   └── decisions.md

When an agent calls recall("src/api"), it receives the full chain of context:

1. Root info.md (Identity)
2. Src info.md (General conventions)
3. API info.md (Specific rules)

This ensures the agent always has the right context for the task at hand.