claw-mem（爪存储器）

ClawMem Skill - OpenClaw Memory Management

Overview

ClawMem is a lightweight memory management system designed for OpenClaw agents. It provides efficient memory storage and retrieval with significant token cost optimization.

Features

• 🎯 3-Tier Retrieval - L0 Index → L1 Timeline → L2 Details
• 👁️ Automatic Lifecycle Monitoring - Intercepts 5 key OpenClaw events
• 💰 Token Optimization - Save 60-80% on token costs
• 🔍 Advanced Search - Keyword/Time/Tag/Session search
• 🗄️ SQLite Storage - High performance, easy deployment

Quick Start

1. Install Dependencies

cd /root/.openclaw/workspace/projects/clawmem
npm install

2. Configure Environment

cp .env.example .env
# Edit .env with your configuration

3. Initialize Database

npm run db:init

4. Use in OpenClaw

import clawMem from './projects/clawmem/src/index.js';

// Store memory
const recordId = clawMem.storeL0({
  category: 'session',
  summary: 'User queried TSLA stock',
  timestamp: Math.floor(Date.now() / 1000)
});

// Retrieve memory
const result = await clawMem.retrieve({
  category: 'session',
  includeTimeline: true,
  limit: 10
});

Usage Examples

Store Memory

// L0 Index (minimal)
clawMem.storeL0({
  category: 'session',
  summary: 'Short summary (< 100 chars)',
  timestamp: Date.now()
});

// L1 Timeline (semantic)
clawMem.storeL1({
  record_id: recordId,
  session_id: 'session_001',
  event_type: 'query',
  semantic_summary: 'Detailed summary (< 500 chars)',
  tags: ['tag1', 'tag2']
});

// L2 Details (full content, on-demand)
clawMem.storeL2({
  record_id: recordId,
  full_content: 'Full content',
  metadata: { key: 'value' }
});

Search Memory

import { memorySearch } from './projects/clawmem/src/index.js';

// Keyword search
const results = memorySearch.searchByKeyword('keyword', {
  category: 'session',
  limit: 10
});

// Time range search
const results = memorySearch.searchByTimeRange({
  start: oneHourAgo,
  end: Date.now() / 1000
});

// Session search
const session = memorySearch.searchBySession('session_001', {
  includeDetails: true
});

// Advanced search
const result = await memorySearch.advancedSearch({
  keyword: 'stock',
  timeRange: { start, end },
  includeDetails: true,
  limit: 10
});

Lifecycle Monitoring

import { lifecycleMonitor } from './projects/clawmem/src/index.js';

// Start monitoring
lifecycleMonitor.start();

// Intercept events
lifecycleMonitor.intercept('tool.call', {
  tool_name: 'yahoo_finance',
  args: { symbol: 'AAPL' }
});

Configuration

Edit .env file:

# Database
DATABASE_PATH=./clawmem.db
DATABASE_WAL_MODE=true

# L0/L1/L2 Limits
L0_MAX_SUMMARY_LENGTH=100
L1_MAX_SUMMARY_LENGTH=500

# Worker Interval
WORKER_INTERVAL_MS=1000

# LLM Configuration
LLM_PROVIDER=openai
LLM_MODEL=gpt-3.5-turbo

API Reference

ClawMemCore

• storeL0(record) - Store minimal index
• storeL1(record) - Store timeline index
• storeL2(record) - Store full details
• retrieve(query) - 3-tier retrieval workflow

MemorySearch

• searchByKeyword(keyword, options) - Keyword search
• searchByTimeRange(timeRange, options) - Time-based search
• searchByTags(tags, options) - Tag-based search
• searchBySession(sessionId, options) - Session search
• advancedSearch(query) - Combined search
• getStats() - Get statistics

LifecycleMonitor

• start() - Start monitoring
• intercept(eventName, payload) - Intercept event

Performance

Token Optimization

Traditional: 100 records × 500 tokens = 50,000 tokens

ClawMem:

• L0: 100 × 25 = 2,500 tokens
• L1: 50 × 125 = 6,250 tokens
• L2: 10 × 500 = 5,000 tokens
• Total: 13,750 tokens (72.5% savings!)

Project Structure

clawmem/
├── src/
│   ├── core/
│   │   ├── retrieval.js
│   │   ├── lifecycle-monitor.js
│   │   └── search.js
│   └── index.js
├── database/
│   └── init.js
├── config/
│   └── loader.js
├── docs/
├── .env.example
├── package.json
└── README.md

License

MIT License

Author

PocketAI for Leo - OpenClaw Community

Repository

https://github.com/leohuang8688/clawmem