# M2Wise

> Memory-to-Wisdom Engine - Extract memories from conversations and generate actionable wisdom strategies for AI companion robots. Use when building AI assistants that need long-term memory and personalized wisdom evolution.

- Skill: `zengyi-thinking/m2wise` (Agent Skill, multi-file: 25 files)
- Install (CLI): `npx skillmds@latest add zengyi-thinking/m2wise`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zengyi-thinking/m2wise/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: zengyi-thinking (https://skillmd.com/u/zengyi-thinking)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/zengyi-thinking/m2wise

---


# M2Wise - Memory-to-Wisdom Engine

> Enable AI companion robots to grow wisdom from experience

## Overview

M2Wise is a **Memory-to-Wisdom Engine** for long-term memory and wisdom evolution in AI companion robots. It extracts memory from raw conversation data and further abstracts them into migratable, verifiable, and versionable wisdom strategies.

## When to Use This Skill

Use M2Wise when you need:
- Long-term memory management for AI companions
- Extract user preferences, facts, and commitments from conversations
- Generate actionable wisdom from accumulated memories
- Track and evolve wisdom effectiveness over time

## Core Features

### 1. Memory Subsystem
- **Preference Memory**: Extract explicit user preferences (e.g., "I prefer concise Chinese technical answers")
- **Fact Memory**: Record user's identity, background, and other factual information
- **Commitment Tracking**: Track user commitments and promises
- **Affective Trace**: Track user emotional changes over time

### 2. Wisdom Subsystem
- **Wisdom Generation**: Automatically generate actionable wisdom from memory clusters
- **Counterexample Mining**: Identify boundary cases where wisdom doesn't apply
- **Confidence Tracking**: Automatically track wisdom effectiveness
- **Self-Evolution**: Auto-adjust wisdom based on hit rates

### 3. Three-Phase Evolution Model
```
Online (Online Phase) → Sleep (Consolidation Phase) → Dream (Verification Phase)
       ↓                        ↓                           ↓
   Real-time Interaction   Memory Extraction + Clustering   Counterexample Mining + Verification & Publishing
```

## Quick Start

```python
from m2wise import M2Wise, M2WiseConfig

# Initialize
config = M2WiseConfig(data_dir="./m2wise_data")
engine = M2Wise(config=config)

# Add user conversation
engine.add(
    [{"role": "user", "content": "I prefer Chinese answers for technical questions"}],
    user_id="alice"
)

# Search memories and wisdom
bundle = engine.search("How to answer technical questions?", user_id="alice")
print(bundle.as_prompt())

# Generate wisdom (Sleep phase)
sleep_report = engine.sleep(user_id="alice")
print(f"Generated {sleep_report.drafts_created} wisdom drafts")

# Verify and publish (Dream phase)
dream_report = engine.dream(user_id="alice")
print(f"Published {dream_report.published} wisdoms")
```

## Technical Features

| Component | Features |
|-----------|----------|
| Memory Extraction | Supports preference, fact, commitment, and more types |
| Similarity Calculation | Jaccard/Cosine/Levenshtein multi-strategy fusion |
| Confidence Evaluation | Time decay + evidence weight + hit rate weighting |
| Caching Strategy | LRU cache + TTL expiration, ~300x speedup |

## Adapter Support

Compatible with multiple external memory systems:
- mem0 compatible
- Letta compatible
- Anthropic message format

## Use Cases

- AI companion robots
- Personalized assistants
- Long-term user profiling systems
- Intelligent customer service

## Installation

```bash
pip install m2wise
```

## Version

Current Version: 1.0.0

## License

Apache-2.0