Memory Architecture Overview

OpenClaw's memory system enables agents to maintain context across conversations and sessions. Unlike stateless chatbots, OpenClaw remembers user preferences, learned patterns, and historical interactions.

Short-term Memory

Short-term memory maintains the current conversation context:

class ConversationMemory {
  private messages: Message[] = [];
  private tokenLimit: number = 8000;
  
  add(message: Message) {
    this.messages.push(message);
    this.enforceLimit();
  }
  
  private enforceLimit() {
    while (this.tokenCount() > this.tokenLimit) {
      this.messages.shift(); // Remove oldest
    }
  }
}

Long-term Memory

Vector Storage

Semantic search for relevant past information:

// Store a memory
await memory.store({
  content: "User prefers dark mode",
  type: "preference",
  timestamp: Date.now()
});

// Recall relevant memories
const related = await memory.recall("theme settings", 5);
// Returns top 5 semantically similar memories

Fact Storage

Structured key-value storage for known facts:

// Store explicit facts
memory.facts.set("user.timezone", "America/Los_Angeles");
memory.facts.set("user.name", "Alice");

// Retrieve facts
const tz = memory.facts.get("user.timezone");

Memory Persistence

Configure where memories are stored:

// config.js
module.exports = {
  memory: {
    shortTerm: {
      maxTokens: 8000,
      persistence: false
    },
    longTerm: {
      vectorStore: 'local', // or 'pinecone', 'weaviate'
      storagePath: './data/memory',
      embeddingModel: 'openai'
    }
  }
}

GitHub Integration for Memory

Use GitHub as a persistent memory backend:

// Store daily summaries
await github.createFile({
  repo: 'my-agent-memory',
  path: `summaries/${date}.md`,
  content: dailySummary
});

// Retrieve past memories
const history = await github.getContents({
  repo: 'my-agent-memory',
  path: 'summaries/'
});

Best Practices

Prune regularly: Remove outdated memories
Categorize: Use memory types for efficient retrieval
Summarize: Condense long conversations into summaries
Privacy: Be mindful of sensitive information storage

Conclusion

Effective memory management transforms OpenClaw from a stateless tool into a personalized assistant that learns and adapts to your needs over time.