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

AI agent memory management is the #1 pain point in 2026. Your OpenClaw agent wakes up fresh each session, relying on manually-written memory files that quickly become stale. Second Brain v2 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
• ✅ Automate everything with nightly pipeline
• ✅ Integrate with qmd for enhanced search

pip install anthropic pyyaml

export ANTHROPIC_API_KEY="your_key_here"

Or configure in OpenClaw auth profiles.

cd ~/workspace/skills/second-brain
python3 scripts/search.py "decisions about model choice"

python3 scripts/stats.py

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

# Run live
python3 scripts/pipeline.py --apply

• Natural language queries over all memory files
• Hybrid search: qmd semantic + BM25 fallback
• Smart relevance ranking with memory scores
• Date range filtering and result highlighting

python3 scripts/search.py "when did we decide on GPT-4.1?"
python3 scripts/search.py --date-range "2026-02-01,2026-02-14" "Discord features"

• Automatically distills daily memories into MEMORY.md
• AI-powered deduplication and merging
• Smart categorization and importance ranking
• Safe dry-run mode with preview

python3 scripts/consolidate.py --dry-run          # Preview changes
python3 scripts/consolidate.py --apply           # Apply changes
python3 scripts/consolidate.py --interactive     # Confirm each section

• Relevance scoring with configurable decay
• Category weights (decisions > facts > transient)
• Frequency boosting for repeated mentions
• Entity importance calculation

python3 scripts/score.py --reindex              # Rebuild scoring index
python3 scripts/score.py --show-top 10         # Top memories
python3 scripts/score.py --decay 7             # Apply time decay

• Structured JSON + Markdown output
• Entity extraction (people, projects, technologies)
• Smart chunking for large sessions
• Better categorization and context

python3 scripts/extract_v2.py --date today     # Extract today's sessions
python3 scripts/extract_v2.py --structured session.jsonl  # Full extraction

• Fully automated memory processing
• Extract → Score → Index → Consolidate → Report
• Configurable thresholds and safety checks
• Comprehensive logging and error handling

python3 scripts/pipeline.py --dry-run          # Test pipeline
python3 scripts/pipeline.py --apply            # Run live

• Memory growth tracking and coverage analysis
• Content statistics and category distributions
• Pipeline health monitoring
• Issue detection and recommendations

python3 scripts/stats.py                       # Health overview
python3 scripts/stats.py --detailed            # Full statistics
python3 scripts/stats.py --issues-only         # Just problems

• Build relationship graphs from memories
• Connect people, projects, decisions, technologies
• Query entity connections and explore relationships
• Export graph data for visualization

python3 scripts/graph.py --build               # Build entity graph
python3 scripts/graph.py --query "DpuDebugAgent"  # Explore connections
python3 scripts/graph.py --stats               # Graph statistics

Edit config.yaml to customize behavior:

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

consolidation:
  interval_days: 7
  dry_run: true            # Safety first!
  min_score: 0.5

scoring:
  decay_half_life_days: 14
  category_weights:
    decisions: 1.5         # Most important
    preferences: 1.3
    facts: 1.0
    action_items: 1.2
    transient: 0.5         # Least important

search:
  use_qmd: true           # Use qmd if available
  fallback: bm25          # Pure Python fallback
  max_results: 10

pipeline:
  extract_recent: true
  update_qmd: true
  consolidate: false      # Require explicit enable
  generate_report: true

~/workspace/skills/second-brain/
├── SKILL.md              # OpenClaw skill documentation  
├── README.md             # This file
├── config.yaml           # Configuration
├── scripts/              # Main functionality
│   ├── search.py         # 🔍 Semantic search
│   ├── consolidate.py    # 🧠 Auto-consolidation  
│   ├── score.py          # 📈 Memory scoring
│   ├── extract_v2.py     # 🔬 Enhanced extraction
│   ├── pipeline.py       # 🔄 Nightly automation
│   ├── stats.py          # 📊 Health dashboard
│   ├── graph.py          # 🕸️ Entity relationships
│   └── utils/            # Core utilities
│       ├── memory_io.py  # File operations
│       ├── scoring.py    # Scoring algorithms
│       ├── bm25.py       # Pure Python search
│       └── qmd_bridge.py # qmd integration
├── tests/                # Unit tests
└── examples/             # Usage examples

Add to your OpenClaw agent's cron config:

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

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

Works with standard OpenClaw memory layout:

~/workspace/memory/
├── 2026-02-01.md         # Daily memories (auto-generated)
├── 2026-02-02.md
├── ...
├── .memory-index.json    # Scoring index (auto-generated)
└── .entity-graph.json    # Entity relationships (auto-generated)

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

Run the test suite:

cd ~/workspace/skills/second-brain
python3 -m pytest tests/ -v

Or run individual test modules:

python3 tests/test_scoring.py
python3 tests/test_search.py  
python3 tests/test_consolidation.py

# Rebuild search index
python3 scripts/score.py --reindex

# Update qmd if available
qmd update && qmd embed

# Check API key
python3 -c "from scripts.utils.memory_io import get_api_key; print('API key OK')"

# Run dry-run first
python3 scripts/consolidate.py --dry-run

# Check what would run
python3 scripts/pipeline.py --dry-run --verbose

# Check recent logs
cat ~/workspace/memory/.pipeline-log.json | tail -n 20

• Memory files: Optimized for up to 10,000 daily files
• Search: qmd provides 10x faster search than BM25 fallback
• Consolidation: Processes last 7 days by default (configurable)
• Entity graph: Rebuild weekly max (expensive operation)
• Pipeline runtime: Typically 30-60 seconds for full run

1. Update config.yaml consolidation categories
2. Extend extraction prompts in extract_v2.py
3. Add category weights in scoring config
4. Update consolidation prompt in consolidate.py

1. Add new search methods in utils/qmd_bridge.py
2. Extend BM25 features in utils/bm25.py
3. Update search ranking in search.py

1. Modify scoring algorithms in utils/scoring.py
2. Add new score factors in calculate_memory_score()
3. Update score distribution analysis in stats.py

MIT License - Use freely in your OpenClaw setups.

This is an OpenClaw skill package. Improvements welcome:

1. Better entity extraction - More accurate NLP for relationships
2. Enhanced search - Semantic similarity, question answering
3. Smarter consolidation - Better duplicate detection, importance ranking
4. Visualization - Graph visualization, timeline views
5. Integration - More OpenClaw framework features

# Morning: Check memory health
python3 scripts/stats.py

# Search for specific information
python3 scripts/search.py "decisions about the Discord bot"

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

# Update scores and consolidate
python3 scripts/score.py --update-only
python3 scripts/consolidate.py --dry-run

# Evening: Run full pipeline
python3 scripts/pipeline.py --apply

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

# Explore project connections
python3 scripts/graph.py --query "second-brain"
python3 scripts/graph.py --query "DpuDebugAgent" 

# Export for analysis
python3 scripts/graph.py --export research_graph.json
python3 scripts/stats.py --format json --export memory_stats.json

# Check what's in memory files
python3 scripts/search.py --verbose "anything" 

# See scoring details
python3 scripts/score.py --show-top 20

# Test pipeline components
python3 scripts/pipeline.py --dry-run --verbose --skip-consolidate

Ready to upgrade your AI agent's memory? Start with python3 scripts/stats.py to see your current memory health! 🧠✨