Memory Enhancement Engine

记忆增加引擎 — Persistent memory engine with hotness-prioritized semantic recall.

Memories that are recalled more often appear first — not just keyword matches.

Quick Start

# Node.js API
const { MemorySystem } = require('./memory-enhancement-engine/memory.js');
const mem = new MemorySystem();

# CLI tool (view / search / compact)
node memory-enhancement-engine/memo.cjs status
node memory-enhancement-engine/memo.cjs query "something to find"

Core Usage

const { MemorySystem } = require('./memory-enhancement-engine/memory.js');

// Create engine (stores to MEMORY_STORE.json automatically)
const mem = new MemorySystem({ decayHalfLifeHours: 2 });

// Write a memory
mem.add({
  content: "Paris is the capital of France.",
  importance: 1.5,
  metadata: { tags: ["geography", "fact"] }
});

// Semantic search
const results = mem.query("France capital");
console.log(results);

// Get recent memories
const recent = mem.recent(10);

// Compact old memories (summarize low-importance clusters)
mem.compact(5, 0.3);

API

Scoring Formula

score = similarity × 0.5 + recency × 0.3 + hotness × 0.2

Where hotness = log(1 + access_count) × exp(-time_delta / 86400)

Features

• Hotness-Prioritized Recall — Frequently accessed memories get boosted scores
• Semantic Search — Bigram overlap + character-level similarity
• Importance Weighting — 0.0 (trivial) to 2.0 (critical)
• Time Decay — Half-life configurable (default 2 hours)
• Auto-Prune — Beyond 1000 entries, least important are pruned
• Auto-Compaction — Merge low-importance groups into summaries
• No Server Needed — Direct Node.js require, stores to local JSON

Installation

File Structure

memory-enhancement-engine/
├── SKILL.md
├── memory.js            # Core engine
├── memo.cjs             # CLI tool
├── package.json
├── assets/
│   └── icon.svg
├── references/
│   ├── API_SPEC.md
│   └── USE_GUIDE.md
└── scripts/
    ├── init-memory.mjs  # One-time migration
    └── test-client.js

License

MIT