andrew-memory（安德鲁记忆）

Andrew Memory Layer

A product-grade semantic memory layer for AI agents, powered by LanceDB.

Overview

This skill provides long-term memory capabilities for AI agents using a local LanceDB vector database. It enables semantic search, Core Identity management, and conversation distillation.

Features

• Semantic Memory Search - Find relevant memories using natural language queries
• Core Identity Injection - Maintain consistent persona across sessions
• Conversation Distillation - Automatically extract memory atoms from conversations
• Dual LLM Support - Works with MiniMax API (cloud) or Ollama (local)
• Rich Metadata - Tracks importance, confidence, and reuse count for each memory

Tools

Configuration

{
  "plugins": {
    "entries": {
      "andrew-memory": {
        "enabled": true,
        "config": {
          "dataDir": "~/.andrew-memory/data",
          "llmMode": "api",
          "localLlmUrl": "http://localhost:11434"
        }
      }
    }
  }
}

Config Options

Requirements

• OpenClaw
• Node.js >= 22
• LanceDB (auto-installed)
• MiniMax API key (if using cloud mode): set MINIMAX_API_KEY env var

Memory Types

• preference - User preferences and habits
• fact - Factual information about the user or world
• rule - Executable rules and guidelines
• experience - Past experiences with success/failure outcomes
• thought - Thoughts, observations, insights
• distilled - Auto-extracted from conversations
• general - Default general-purpose memory

Usage Example

User: Remember that I prefer short responses in the morning.
→ andrew_memory_add: { text: "User prefers short responses in the morning", memoryType: "preference" }

User: What did I say about my communication preferences?
→ andrew_memory_search: { query: "communication preferences morning responses" }