Mempalace

A community open-source plugin that integrates MemPalace's "lifetime memory" (L0-L3 memory stack, AAAK compression, auto context saving) into the OpenCode terminal assistant.

This plugin ensures your AI assistant has a long-term memory across sessions by seamlessly hooking into OpenCode's lifecycle events to fetch, inject, and save contexts related to your specific workspace.

🌟 Features

• Zero-Config Auto-Initialization: Opens a new folder? The plugin automatically initializes a MemPalace database for it in the background.
• Auto-Injection (Wake-up): Automatically wakes up MemPalace on session initialization to inject L0 (Global Identity) and L1 (Critical Facts) directly into the AI's System Prompt.
• Pre-Compaction Rescue: Adds your core memory context back right before OpenCode compresses the conversation, ensuring crucial details are never lost.
• Silent Background Mining: Quietly exports and saves your conversational history into your MemPalace database as you chat, preserving decisions for future usage without spending extra tokens on MCP tool calls.
• Crash Safety & Idle Auto-Save: Never lose your context, even if you close the terminal early!
  • Idle Auto-Save: If your session is deleted or you simply stop chatting (idle), any un-saved messages are softly mined in the background.
  • Crash Safety: If you force-quit the terminal (Ctrl+C), the plugin intercepts the exit signal and performs a synchronous emergency save.

📋 Requirements

• OpenCode AI Terminal
• Python 3.9+
• MemPalace installed globally (pip install mempalace or python3 -m pip install mempalace)

🚀 Installation

1. Install the official MemPalace CLI on your system:

   python3 -m pip install mempalace
   

2. Add this plugin to your OpenCode configuration. Open ~/.config/opencode/opencode.json (or your project's local opencode.json) and add the package name to the plugins array:

   {
     "$schema": "https://opencode.ai/config.json",
     "plugin": ["opencode-plugin-mempalace"]
   }
   

⚙️ Configuration (Optional)

You can pass configuration options to the plugin to customize its behavior. Currently supported options:

• threshold (default: 15): The number of chat messages required before the plugin triggers a background auto-save (mining) of your conversation.

{
  "plugin": [["opencode-plugin-mempalace", { "threshold": 20 }]]
}

🛠️ How It Works

• The plugin wraps the mempalace CLI via the execa package.
• It acts as the "subconscious" of your AI, rather than just an active tool.
• Hooks used:
  • experimental.chat.system.transform: Injects memory.
  • experimental.session.compacting: Rescues memory from truncation.
  • chat.message: Tracks conversation length and triggers background mining.
  • event: Listens for session.idle and session.deleted for soft-exit saving.
  • process.on('exit' | 'SIGINT' | 'SIGTERM'): Intercepts hard process exits for emergency synchronous saving.
• Workspace Isolation: It infers the wing name intelligently from the workspace path (e.g. /projects/my-app -> wing_my-app). Your memory stays isolated per project!

🧑‍💻 Development

This project is built with TypeScript (ESM) and relies on a strict TDD approach.

npm install
npm run lint
npm run test
npm run build

📄 License

MIT