Second Brain v2 — OpenClaw Memory Intelligence Skill

Transform your agent from forgetful to brilliant with automated memory management.

Overview

Second Brain v2 is an OpenClaw skill that solves the #1 AI agent problem in 2026: memory loss between sessions. Your agent wakes up fresh every time, relying on manually-written memory files that quickly become stale. This skill automates the entire memory lifecycle:

• Extract memories from session logs using Claude
• Search memories with natural language queries
• Consolidate daily memories into long-term storage automatically
• Score memories by relevance with smart decay algorithms
• Track memory health with detailed statistics
• Graph entity relationships across all memories

Key Features

🔍 Semantic Memory Search

Query your memories naturally: remember search "when did we decide on GPT-4.1?"

🧠 Auto-Consolidation

Automatically distills daily memories into your MEMORY.md file with smart deduplication

📊 Memory Scoring & Decay

Relevance scoring with configurable decay (decisions matter more than transient notes)

🔄 Nightly Pipeline

Fully automated memory processing that runs in background via cron

📈 Health Dashboard

Track memory growth, coverage gaps, and consolidation freshness

🕸️ Entity Graph

Visualize relationships between people, projects, and decisions

Installation

# Install the skill (if using OpenClaw skill manager)
openclaw skill install second-brain

# Or clone directly
git clone <repo> ~/workspace/skills/second-brain

Dependencies

• Python 3.12+
• anthropic SDK (pip install anthropic)
• PyYAML (pip install pyyaml)
• qmd (optional, for enhanced search)

Quick Start

1. Configure Your API Key

Set your Anthropic API key:

export ANTHROPIC_API_KEY="your_key_here"

Or configure it in OpenClaw auth profiles at ~/.openclaw/agents/main/agent/auth-profiles.json.

2. Search Existing Memories

python3 scripts/search.py "DPU debugging decisions"
python3 scripts/search.py --date-range "2026-02-01,2026-02-14" "Discord bot features"

3. Run Memory Health Check

python3 scripts/stats.py

4. Extract Memories from Recent Sessions

# Extract from today's sessions
python3 scripts/extract_v2.py --date today

# Extract specific session
python3 scripts/extract_v2.py ~/path/to/session.jsonl

5. Auto-Consolidate Recent Memories

# Dry run (see proposed changes)
python3 scripts/consolidate.py --days 7 --dry-run

# Apply consolidation
python3 scripts/consolidate.py --days 7 --apply

6. Set Up Nightly Pipeline

# Test the full pipeline
python3 scripts/pipeline.py --dry-run

# Schedule via OpenClaw cron (recommended)
# Add to your agent's cron config

Commands Reference

Memory Search

# Natural language search
python3 scripts/search.py "your query here"

# Advanced options
python3 scripts/search.py \
    --date-range "2026-02-01,2026-02-14" \
    --limit 5 \
    --format json \
    "decision about model choice"

Memory Extraction

# Extract from date
python3 scripts/extract_v2.py --date 2026-02-14

# Extract specific session
python3 scripts/extract_v2.py session_123.jsonl

# Enhanced output with entities
python3 scripts/extract_v2.py --structured session.jsonl

Memory Consolidation

# Dry run (preview only)
python3 scripts/consolidate.py --dry-run

# Consolidate last 7 days
python3 scripts/consolidate.py --days 7 --apply

# Interactive mode
python3 scripts/consolidate.py --interactive

Memory Scoring

# Score all memories
python3 scripts/score.py --reindex

# Show top-scored memories
python3 scripts/score.py --show-top 20

# Update scores only
python3 scripts/score.py --update-only

Health & Statistics

# Memory health dashboard
python3 scripts/stats.py

# Detailed breakdown
python3 scripts/stats.py --detailed

# Export for external tools
python3 scripts/stats.py --format json

Entity Graph

# Build entity relationship graph
python3 scripts/graph.py --build

# Query graph
python3 scripts/graph.py --query "DpuDebugAgent"

# Export graph data
python3 scripts/graph.py --export graph.json

Nightly Pipeline

# Full pipeline (recommended for cron)
python3 scripts/pipeline.py

# Pipeline with custom config
python3 scripts/pipeline.py --config my_config.yaml

# Dry run mode
python3 scripts/pipeline.py --dry-run

Configuration

Default config is in config.yaml. Override by creating your own config file:

memory_dir: ~/workspace/memory
session_dir: ~/.openclaw/agents/main/sessions
memory_md: ~/workspace/MEMORY.md

consolidation:
  interval_days: 7
  dry_run: false
  min_score: 0.5
  backup_memory_md: true

scoring:
  decay_half_life_days: 14
  category_weights:
    decisions: 1.5
    preferences: 1.3
    facts: 1.0
    action_items: 1.2
    transient: 0.5

search:
  use_qmd: true
  fallback: bm25
  max_results: 10
  highlight_matches: true

pipeline:
  extract_recent: true
  update_qmd: true
  generate_report: true
  consolidate_threshold: 5  # Min new memories before consolidation

Integration with OpenClaw

Cron Integration

Add to your OpenClaw agent's cron configuration:

jobs:
  - name: "memory-pipeline"
    schedule: "0 2 * * *"  # 2 AM daily
    command: "python3 ~/workspace/skills/second-brain/scripts/pipeline.py"
    enabled: true

Using in Agent Sessions

The skill is designed to work seamlessly with OpenClaw agents. Commands can be called directly from sessions:

# In your agent code
exec(command="python3 ~/workspace/skills/second-brain/scripts/search.py 'recent decisions'")

File Structure

~/workspace/skills/second-brain/
├── SKILL.md              # This file
├── README.md             # Full documentation
├── config.yaml           # Default configuration
├── scripts/
│   ├── search.py         # Semantic memory search
│   ├── consolidate.py    # Auto-consolidation engine
│   ├── score.py          # Memory scoring & decay
│   ├── extract_v2.py     # Enhanced extraction
│   ├── pipeline.py       # Nightly automation
│   ├── stats.py          # Health dashboard  
│   ├── graph.py          # Entity relationships
│   ├── extract.py        # Original extractor (legacy)
│   ├── digest.py         # Original digester (legacy)
│   ├── weekly.py         # Weekly analysis (legacy)
│   └── utils/
│       ├── memory_io.py  # Memory file operations
│       ├── scoring.py    # Scoring algorithms
│       ├── qmd_bridge.py # qmd search integration
│       └── bm25.py       # Pure Python BM25 implementation
├── tests/
│   ├── test_search.py
│   ├── test_scoring.py
│   ├── test_consolidation.py
│   └── test_pipeline.py
└── examples/
    ├── sample_config.yaml
    └── cron_setup.md

Memory Files Layout

The skill works with OpenClaw's standard memory organization:

~/workspace/memory/
├── 2026-02-01.md         # Daily memory files
├── 2026-02-02.md
├── ...
├── .memory-index.json    # Scored memory index
├── .pipeline-log.json    # Pipeline execution log
└── .entity-graph.json    # Entity relationships

Plus your main memory file:

~/workspace/MEMORY.md     # Long-term consolidated memories

Troubleshooting

Search Returns No Results

1. Check if memory files exist in ~/workspace/memory/
2. Try rebuilding search index: python3 scripts/score.py --reindex
3. If using qmd, run: qmd update && qmd embed

Consolidation Not Working

1. Check API key is configured: python3 -c "from scripts.utils.memory_io import get_api_key; print('OK')"
2. Run with --dry-run first to see what would be consolidated
3. Check minimum score threshold in config

Pipeline Failing

1. Run python3 scripts/pipeline.py --dry-run to see what would happen
2. Check logs in ~/workspace/memory/.pipeline-log.json
3. Verify all dependencies are installed

Permission Errors

1. Ensure OpenClaw agent has write access to ~/workspace/memory/
2. Check that MEMORY.md is writable
3. Backup files before running consolidation with --apply

Performance Notes

• Memory search is optimized for up to 10,000 daily memory files
• Consolidation processes last 7 days by default (configurable)
• Entity graph rebuilds are expensive - run weekly max
• qmd integration provides much faster search than BM25 fallback

Contributing

This is an OpenClaw skill. Improvements welcome:

1. Add new memory categories
2. Improve scoring algorithms
3. Better entity extraction
4. Enhanced search capabilities

License

MIT License - Use freely in your OpenClaw setups.