AI Bot Memory System Guide

Overview

A good AI assistant remembers you. It knows your name, preferences, past conversations, and context. This guide explains how to configure memory for OpenClaw and OpenClaw so your bot becomes truly personal.

How Memory Works

New Message
    ↓
[Load relevant memories]
    ↓
[Add to context]
    ↓
[AI processes with context]
    ↓
[Save new memories]
    ↓
Response

Memory Types

1. Conversation Context

Short-term memory of the current conversation:

# Keep last N messages in context
MAX_CONTEXT_MESSAGES=30

# Context timeout (minutes of inactivity)
CONTEXT_TIMEOUT_MINUTES=60

Example:

You: My name is Alex
Bot: Nice to meet you, Alex!
You: What's my name?
Bot: Your name is Alex.  ← Uses conversation context

2. User Preferences

Long-term storage of user-specific settings:

# Enable preference storage
USER_PREFERENCES_ENABLED=true
PREFERENCES_FILE=/opt/openclaw/data/preferences.json

Stored preferences:

Name, timezone, language
Preferred response style
Notification settings
Integration credentials

3. Daily Memory Files

Automatic daily logs of interactions:

# Enable daily memory
DAILY_MEMORY_ENABLED=true
MEMORY_PATH=/opt/openclaw/memory
MEMORY_FORMAT=markdown

Creates files like:

/memory/2026-01-27.md
/memory/2026-01-28.md

4. Semantic Memory

AI-searchable knowledge about the user:

# Enable semantic memory
SEMANTIC_MEMORY_ENABLED=true
EMBEDDING_MODEL=text-embedding-3-small
MEMORY_SEARCH_RESULTS=5

Configuration Guide

Basic Setup

# Memory Settings
MEMORY_ENABLED=true
MEMORY_PATH=/opt/openclaw/memory

# Conversation context
MAX_CONTEXT_MESSAGES=30
CONTEXT_TIMEOUT_MINUTES=60

# Long-term memory
LONG_TERM_MEMORY=true
MEMORY_RETENTION_DAYS=90

Memory File Structure

/opt/openclaw/
├── memory/
│   ├── daily/
│   │   ├── 2026-01-27.md
│   │   └── 2026-01-28.md
│   ├── users/
│   │   └── user_123456.json
│   ├── preferences.json
│   └── facts.json

Daily Memory Format

# Memory - January 27, 2026

## Conversations

### 09:15 - Morning briefing request
User asked for daily briefing. Mentioned having a busy day with 3 meetings.

### 14:30 - Task management
User completed "Send proposal" task. Asked to reschedule "Client call" to tomorrow.

### 18:45 - Evening check-in
User asked about weather for tomorrow (planning outdoor activity).

## Facts Learned
- User has meeting with Sarah on Wednesday
- User prefers morning briefings at 7:30 AM
- User is planning a trip next month

## Preferences Updated
- Changed notification time from 8:00 to 7:30

Making Memory Work

Explicit Memory Commands

You: "Remember that I prefer metric units" Bot: "Noted! I'll use metric units in all my responses."

You: "What do you know about me?" Bot: "Here's what I remember:

Name: Alex
Timezone: Europe/London
Prefers metric units
Morning briefing at 7:30 AM
Working on 'Website Redesign' project"

You: "Forget my phone number" Bot: "Done. I've removed your phone number from my memory."

Implicit Memory

The bot learns from conversations:

Conversation 1: You: "I'm vegetarian" Bot: "Good to know!"

Later: You: "Suggest a restaurant for dinner" Bot: "Here are some vegetarian-friendly options near you..." ← Remembered preference

Memory Recall

You: "What did we discuss about the project last week?" Bot: "Last week (Jan 20-24) we discussed:

Jan 20: Initial project scope
Jan 22: Budget concerns
Jan 24: Timeline adjusted to March deadline

Would you like more details on any of these?"

System Prompt for Memory

Add memory context to your system prompt:

SYSTEM_PROMPT="You are a personal AI assistant with memory capabilities.

You remember:
- User's name and preferences
- Past conversations and decisions
- Important dates and commitments
- User's projects and goals

When responding:
- Reference past conversations when relevant
- Use remembered preferences
- Remind about upcoming commitments
- Build on previous discussions

Memory instructions:
- When user shares personal info, store it
- When user makes decisions, remember them
- When user has preferences, apply them consistently
- If unsure about something previously discussed, acknowledge the gap"

Advanced Memory Features

Memory Search

Find relevant past information:

async function searchMemory(query, userId) {
  // Get embeddings for query
  const queryEmbedding = await getEmbedding(query);

  // Search memory vectors
  const results = await vectorStore.search({
    vector: queryEmbedding,
    filter: { userId },
    topK: 5
  });

  return results.map(r => r.content);
}

You: "What did I decide about the logo?" Bot: Searches memory, finds: "On Jan 15, you decided to go with the blue logo variant with the modern font."

Memory Importance Ranking

Not all information is equally important:

async function assessImportance(message) {
  const response = await claude.messages.create({
    model: 'claude-3-5-sonnet-20241022',
    messages: [{
      role: 'user',
      content: `Rate the importance of remembering this information (1-10):

"${message}"

Consider: Is this a preference? A decision? A fact about the user? Temporary or permanent?

Return JSON: {"importance": N, "category": "preference|fact|decision|temporary", "summary": "brief summary"}`
    }]
  });

  return JSON.parse(response.content[0].text);
}

Memory Consolidation

Periodically summarize and consolidate memories:

// Run weekly
async function consolidateMemory(userId) {
  const weekMemories = await getWeekMemories(userId);

  const summary = await claude.messages.create({
    model: 'claude-3-5-sonnet-20241022',
    messages: [{
      role: 'user',
      content: `Consolidate these memories into key facts and preferences:

${weekMemories.map(m => `- ${m}`).join('\n')}

Return the most important information to remember long-term.`
    }]
  });

  await saveLongTermMemory(userId, summary);
  await archiveWeekMemories(userId);
}

Privacy and Security

Data Retention

# Auto-delete old memories
MEMORY_RETENTION_DAYS=90

# Keep important facts longer
FACTS_RETENTION_DAYS=365

# Never store sensitive data
EXCLUDE_FROM_MEMORY=password,credit card,ssn,api key

User Control

Let users manage their memory:

You: "/memory" Bot: "Memory options:

/memory show - View what I remember
/memory clear - Clear all memory
/memory delete [topic] - Delete specific memory
/memory export - Download your data"

Encryption

Encrypt memory at rest:

MEMORY_ENCRYPTION=true
MEMORY_ENCRYPTION_KEY=your-encryption-key

Troubleshooting

Bot doesn't remember

Check memory is enabled: MEMORY_ENABLED=true
Verify memory path is writable
Check context timeout isn't too short

Memory too short

Increase context:

MAX_CONTEXT_MESSAGES=50
CONTEXT_TIMEOUT_MINUTES=120

Irrelevant memories recalled

Improve memory search:

MEMORY_RELEVANCE_THRESHOLD=0.7
MEMORY_SEARCH_RESULTS=3

Memory file growing too large

Enable consolidation:

MEMORY_CONSOLIDATION=weekly
MEMORY_MAX_SIZE_MB=100

Best Practices

Do's

✅ Let users see what's remembered
✅ Allow memory deletion
✅ Consolidate old memories
✅ Separate by importance
✅ Encrypt sensitive data

Don'ts

❌ Store passwords or credentials
❌ Keep memories forever without cleanup
❌ Make memory impossible to delete
❌ Share memory between users

Related Guides

Need Help?

Memory configuration affects the entire bot experience. Our setup service includes memory system configuration tailored to your use case.