Open Source

Teaching AI to Remember

A file-based memory system for AI agents that forget between sessions. Named after the forgetful fish — because your AI is brilliant, but it's also a goldfish.

# Your agent reads this every session
workspace/
├── SOUL.md          # Who the agent is
├── USER.md          # Who you are
├── MEMORY.md        # Long-term memory
│
├── state/           # Check every session
│   ├── ACTIVE.md    # Current task (exact words)
│   ├── HOLD.md      # Blocked items
│   └── DECISIONS.md # Recent choices
│
└── memory/
    └── YYYY-MM-DD.md

🐠 The Dory-Proof Pattern

When the user gives a task, immediately write their exact words to a file. Then interpret. Then act.

Not a paraphrase. Not a summary. The exact words. Because interpretation drifts, but quotes don't.

📁 Core Principles

Text > Brain

Files ARE memory. If it's not written down, it doesn't exist.

Structure Prevents Chaos

HOLD overrides ACTIVE overrides DECISIONS. Clear hierarchy.

Exact Words Matter

Capture user's exact words before interpreting. Prevents drift.

Boot Sequence

Every session reads state files first. Continuity guaranteed.

Level Up with QMD ✓ Built-in

OpenClaw 2026.2.6+ includes QMD — a local semantic search engine. Auto-indexes your workspace every 5 minutes. Searches by meaning, not just keywords. Session transcripts included.

// Enable in openclaw.json
{
  "memory": {
    "backend": "qmd",
    "qmd": {
      "update": { "interval": "5m" },
      "sessions": { "enabled": true }
    }
  }
}

The Dory-Proof pattern remains the foundation. QMD adds intelligent retrieval on top.

"The Dory Problem isn't about AI having bad memory. It's about us expecting AI to work like humans when it doesn't."

— From the original article

Get Started

Install the skill, copy the templates, start remembering.