3-Layer Memory Architecture

The Problem

Every OpenClaw user hits the same wall: your agent forgets things. After compaction, after restarts, after context fills up. Important decisions, user preferences, project context — gone.

The Solution: 3 Layers

Layer 1: Knowledge Graph (/life/areas/)

Entity-based storage for durable facts about people, companies, projects.

life/areas/
├── people/tina/
│   ├── summary.md      # Who is Tina
│   └── items.json      # Atomic facts (timestamped)
├── projects/years/
│   ├── summary.md
│   └── items.json
└── system/crons/
    ├── summary.md
    └── items.json

Facts are never deleted — only superseded. Your agent searches with qmd search or direct file reads.

Layer 2: Daily Notes (memory/YYYY-MM-DD.md)

Raw event log. Everything that happened today. Durable facts get extracted to Layer 1 by a cron job.

Layer 3: Tacit Knowledge (MEMORY.md)

Patterns, preferences, lessons learned. Updated manually or by the agent. This is the agent's long-term intuition.

What's Included

• Complete SKILL.md with setup instructions
• Directory structure templates
• Cron job config for automated fact extraction
• AGENTS.md additions for memory protocols
• Example Knowledge Graph entries
• Migration guide from flat-file memory

Why This Works

This system has been running since January 2026 across 6 agents. 132+ entity folders, 43 auto-generated summaries, zero information loss after 50+ compactions.