Long-Term Memory Management

Systematic management of MEMORY.md and daily memory files for persistent knowledge retention.

Memory Architecture

~/.openclaw/workspace/
├── MEMORY.md              # Long-term curated memory (main)
├── memory/                # Daily memory files
│   ├── 2025-01-20.md
│   ├── 2025-01-21.md
│   └── ...
└── .memory-archive/       # Archived memories
    └── 2025-01/
        ├── consolidated.md
        └── raw/

Quick Commands

# Consolidate recent daily memories
python3 {baseDir}/scripts/memory_manager.py --consolidate --days 7

# Archive old memories
python3 {baseDir}/scripts/memory_manager.py --archive --older-than 30

# Extract key facts from MEMORY.md
python3 {baseDir}/scripts/memory_manager.py --extract-facts

# Search memory history
python3 {baseDir}/scripts/memory_manager.py --search "关键词"

# Generate memory summary
python3 {baseDir}/scripts/memory_manager.py --summary --output memory-summary.md

# Health check
python3 {baseDir}/scripts/memory_manager.py --health

MEMORY.md Management

Structure

MEMORY.md should contain distilled, long-term knowledge:

# MEMORY.md - Long-Term Memory

## User Profile
- Name: ...
- Preferences: ...
- Work patterns: ...

## Key Decisions
- [Date] Decision: Reasoning...

## Important Facts
- Account: location...
- Credentials: stored in...
- Recurring tasks: ...

## Lessons Learned
- Pattern: Insight...

## Active Projects
- Project A: Status, next steps...

## Recurring Context
- Weekly meetings: ...
- Regular reports: ...

Consolidation Workflow

1. Review daily memories (last 7 days)
2. Extract significant items:
   • User preferences mentioned
   • Important decisions made
   • New facts discovered
   • Lessons learned
   • Active project updates
3. Update MEMORY.md with distilled content
4. Archive processed daily files

# Full consolidation workflow
python3 {baseDir}/scripts/memory_manager.py --consolidate --auto-archive

What to Keep in MEMORY.md

Daily Memory Files

Purpose

memory/YYYY-MM-DD.md files capture:

• What happened today
• Important conversations
• Decisions made
• Tasks completed
• Context for future reference

Best Practices

DO:

# 2025-01-20

## Key Events
- User asked about X, decided Y
- Set up new integration Z
- Discovered preference for concise responses

## Decisions
- Use tool X instead of Y for Z task (user preference)

## Pending
- Follow up on ...

DON'T:

# 2025-01-20

Got message. Replied HEARTBEAT_OK.
User said hi. Said hi back.
Time is 3pm.

Automatic Extraction

The system can extract valuable content:

# Extract what matters from daily files
python3 {baseDir}/scripts/memory_manager.py --extract --from "2025-01-20.md"

# Output: List of extractable facts

Archival System

Archive Threshold

Default: Archive daily memories older than 30 days

# Archive old memories
python3 {baseDir}/scripts/memory_manager.py --archive --older-than 30

# Archive with consolidation
python3 {baseDir}/scripts/memory_manager.py --archive --older-than 30 --consolidate-first

Archive Structure

.memory-archive/
├── 2025-01/
│   ├── consolidated.md    # Summary of the month
│   └── raw/               # Original daily files
│       ├── 2025-01-01.md
│       └── ...
└── 2025-02/
    └── ...

Retrieval from Archive

# Search archived memories
python3 {baseDir}/scripts/memory_manager.py --search "keyword" --include-archive

# Retrieve specific archived content
python3 {baseDir}/scripts/memory_manager.py --retrieve "2025-01-15"

Memory Compression

What Gets Compressed

Daily memories contain repetition and noise. Compression extracts:

1. Unique events - Things that happened once
2. Recurring patterns - Things that repeat
3. Key decisions - Choices made
4. Important facts - Persistent information

Compression Rules

# Compress with custom rules
python3 {baseDir}/scripts/memory_manager.py --compress \
  --rules keep-decisions,keep-preferences,keep-facts \
  --remove heartbeets,trivial,transient

Example Compression

Before (daily files, 5000 words):

# 2025-01-20
User asked about API. Looked up docs. Found answer.
User preferred concise response. Noted preference.
...

# 2025-01-21
User asked about API again. Provided concise answer.
User appreciated brevity.
...

After (MEMORY.md, 100 words):

## User Preferences
- Prefers concise responses over detailed explanations

## Knowledge
- API documentation location: ...

## Lessons
- Concise answers are preferred for API questions

Memory Search

Search Commands

# Search all memories
python3 {baseDir}/scripts/memory_manager.py --search "关键词"

# Search specific range
python3 {baseDir}/scripts/memory_manager.py --search "..." --from 2025-01-01 --to 2025-01-31

# Search with context
python3 {baseDir}/scripts/memory_manager.py --search "..." --context 3

# Search archives too
python3 {baseDir}/scripts/memory_manager.py --search "..." --include-archive

Search Output

{
  "query": "关键词",
  "results": [
    {
      "date": "2025-01-20",
      "file": "memory/2025-01-20.md",
      "line": 15,
      "context": "...",
      "relevance": "high"
    }
  ],
  "total": 3
}

Integration with Vector Memory

This skill works alongside vector memory (LanceDB):

Coordination

# Consolidate both systems
python3 {baseDir}/scripts/memory_manager.py --consolidate --sync-vector

# The script will:
# 1. Update MEMORY.md
# 2. Archive old daily files
# 3. Sync key facts to vector memory

Automation

Periodic Consolidation

Add to heartbeat or cron:

# HEARTBEAT.md
- Run memory consolidation weekly
- Archive memories older than 30 days
- Sync to vector memory

Or via cron:

# Weekly consolidation (Sunday 4am)
cron action=add job='{
  "name": "memory-consolidation",
  "schedule": "0 4 * * 0",
  "text": "Consolidate weekly memories: 1) Review memory/ files 2) Update MEMORY.md 3) Archive old files 4) Sync to vector memory"
}'

Automated Extraction

During heartbeats, automatically extract:

python3 {baseDir}/scripts/memory_manager.py --auto-extract --days 1

Memory Health

Health Metrics

Health Check

python3 {baseDir}/scripts/memory_manager.py --health

# Output

{
  "status": "healthy",
  "metrics": {
    "daily_files": 15,
    "memory_md_size": "12KB",
    "last_consolidation": "2025-01-18",
    "archive_coverage": "95%"
  },
  "recommendations": []
}

Cleanup

# Clean up stale content
python3 {baseDir}/scripts/memory_manager.py --cleanup

# Remove duplicates
python3 {baseDir}/scripts/memory_manager.py --dedupe

# Reorganize structure
python3 {baseDir}/scripts/memory_manager.py --reorganize

Best Practices

Writing to MEMORY.md

1. Be concise - Distill, don't copy
2. Use structure - Consistent sections
3. Date important items - When did this become true?
4. Review periodically - Remove outdated info
5. One concept per line - Easy to search

Managing Daily Files

1. Write daily - Capture what matters
2. Skip trivial - No heartbeat logs
3. Link related - Reference other files
4. Mark important - Use clear headings
5. Archive promptly - Don't accumulate

Integration with Other Skills

# Use with self-evolution
python3 {baseDir}/scripts/memory_manager.py --consolidate
python3 ../self-evolution/scripts/evolution.py --analyze --with-memory

# Use with self-improvement
# Log a learning, then consolidate
python3 {baseDir}/scripts/memory_manager.py --extract-from .learnings/LEARNINGS.md

Workflow Examples

Weekly Maintenance

# 1. Check health
python3 {baseDir}/scripts/memory_manager.py --health

# 2. Consolidate recent memories
python3 {baseDir}/scripts/memory_manager.py --consolidate --days 7

# 3. Archive old files
python3 {baseDir}/scripts/memory_manager.py --archive --older-than 30

# 4. Sync to vector memory
python3 {baseDir}/scripts/memory_manager.py --sync-vector

# 5. Generate report
python3 {baseDir}/scripts/memory_manager.py --summary

After Important Session

# Extract key facts immediately
python3 {baseDir}/scripts/memory_manager.py --extract --today

# Update MEMORY.md
python3 {baseDir}/scripts/memory_manager.py --update --section "Key Decisions" --add "..."

Research Task

# Search all memories for context
python3 {baseDir}/scripts/memory_manager.py --search "项目名" --include-archive --context 5

# Export relevant memories
python3 {baseDir}/scripts/memory_manager.py --export "项目名" --output project-context.md

Notes

• MEMORY.md is for main sessions only (not group chats)
• Daily files should be raw logs, not polished documents
• Archiving preserves data, just moves it out of active view
• Consolidation is distillation, not summarization
• The goal: quick recall of important information