system-prompts-and-models-of-ai-tools/.github/Advanced_AI_Agent/prompts/memory_system.md

🧠 Advanced Memory System

Synthesized from Industry Best Practices

Overview

The Advanced Memory System represents the culmination of analyzing memory patterns from leading AI assistants including Cursor, Devin AI, Manus, and others. This system enables persistent learning, context preservation, and natural memory integration across sessions.

---

🎯 Core Principles

1. Natural Integration

• Seamless Citations: Use [[memory:MEMORY_ID]] format for natural memory integration
• Context Preservation: Maintain important context across multiple sessions
• Proactive Storage: Create memories immediately when encountering important information
• Selective Updates: Update or delete memories when contradicted or augmented

2. Multi-Level Context

• User Preferences: Communication style, expertise level, project preferences
• Project Patterns: Code conventions, architecture decisions, dependencies
• Technical Knowledge: Solutions, workarounds, best practices
• Interaction History: Previous decisions, successful approaches, lessons learned

3. Intelligent Management

• Validation: Verify memory accuracy before using
• Cleanup: Remove outdated or incorrect memories
• Contradiction Handling: Delete memories when contradicted by user
• Augmentation: Update memories with new information

---

📋 Memory Types

User Preferences Memory

interface UserPreferences {
  communicationStyle: "detailed" | "concise" | "technical" | "conversational";
  expertiseLevel: "beginner" | "intermediate" | "advanced" | "expert";
  projectPreferences: {
    preferredLanguages: string[];
    codingStyle: "functional" | "objectOriented" | "procedural";
    documentationLevel: "minimal" | "standard" | "comprehensive";
  };
  interactionPatterns: {
    preferredResponseFormat: "summary" | "detailed" | "stepByStep";
    learningStyle: "visual" | "handsOn" | "theoretical";
  };
}

Project Patterns Memory

interface ProjectPatterns {
  codeConventions: {
    namingConventions: Record<string, string>;
    fileStructure: string[];
    importPatterns: string[];
  };
  architectureDecisions: {
    frameworkChoices: Record<string, string>;
    designPatterns: string[];
    dependencyManagement: string;
  };
  technicalDebt: {
    knownIssues: string[];
    plannedImprovements: string[];
    workarounds: Record<string, string>;
  };
}

Technical Knowledge Memory

interface TechnicalKnowledge {
  solutions: {
    problem: string;
    solution: string;
    context: string;
    effectiveness: "high" | "medium" | "low";
  }[];
  bestPractices: {
    category: string;
    practice: string;
    rationale: string;
    examples: string[];
  }[];
  workarounds: {
    issue: string;
    workaround: string;
    permanentSolution?: string;
  }[];
}

Interaction History Memory

interface InteractionHistory {
  decisions: {
    context: string;
    decision: string;
    rationale: string;
    outcome: "successful" | "failed" | "partial";
  }[];
  successfulApproaches: {
    taskType: string;
    approach: string;
    keyFactors: string[];
  }[];
  lessonsLearned: {
    situation: string;
    lesson: string;
    application: string;
  }[];
}

---

🔧 Memory Operations

Memory Creation

// Create a new memory
update_memory({
  title: "User prefers concise responses",
  knowledge_to_store: "User prefers brief, actionable responses over detailed explanations. Focus on key points and next steps.",
  action: "create"
});

Memory Citation

// Use memory in response
"Based on your preference for concise responses [[memory:user_communication_style]], I'll provide the key points directly."

// Natural integration example
"I'll implement the authentication system using JWT tokens [[memory:project_auth_pattern]], following the established patterns in your codebase."

Memory Updates

// Update existing memory
update_memory({
  title: "User prefers concise responses",
  knowledge_to_store: "User prefers concise responses but appreciates detailed explanations for complex technical topics.",
  action: "update",
  existing_knowledge_id: "user_communication_style"
});

Memory Deletion

// Delete contradicted memory
update_memory({
  action: "delete",
  existing_knowledge_id: "outdated_technology_choice"
});

---

🎯 Memory Usage Patterns

1. Context-Aware Responses

// Example: Adapting communication style
if (hasMemory("user_expertise_level") === "beginner") {
  return provideDetailedExplanation();
} else {
  return provideConciseSummary();
}

2. Pattern Recognition

// Example: Recognizing recurring patterns
if (hasMemory("similar_problem_solved")) {
  return applyKnownSolution();
} else {
  return exploreNewApproach();
}

3. Learning Integration

// Example: Learning from previous interactions
if (hasMemory("failed_approach")) {
  return avoidPreviousMistake();
} else {
  return tryProvenMethod();
}

---

📊 Memory Quality Metrics

Accuracy Metrics

• Validation Rate: Percentage of memories verified as accurate
• Contradiction Rate: Frequency of memory contradictions
• Update Frequency: How often memories are updated
• Usage Effectiveness: Impact of memory usage on response quality

Performance Metrics

• Retrieval Speed: Time to access relevant memories
• Context Relevance: Percentage of memories relevant to current context
• Memory Density: Amount of useful information per memory
• Cross-Session Persistence: Memory retention across sessions

---

🔍 Memory Search and Retrieval

Semantic Search

// Search memories by semantic similarity
searchMemories({
  query: "authentication implementation",
  context: "current_project",
  relevanceThreshold: 0.7
});

Context-Aware Retrieval

// Retrieve memories based on current context
getRelevantMemories({
  currentTask: "implement_user_auth",
  userExpertise: "intermediate",
  projectType: "web_application"
});

Pattern Matching

// Find memories matching specific patterns
findPatternMemories({
  pattern: "error_handling",
  technology: "python",
  context: "api_development"
});

---

🛡️ Memory Safety and Ethics

Privacy Protection

• User Consent: Only store memories with implicit or explicit user consent
• Data Minimization: Store only necessary information
• Anonymization: Remove personally identifiable information
• Retention Policies: Automatically expire outdated memories

Bias Mitigation

• Diversity Awareness: Avoid reinforcing existing biases
• Balanced Learning: Learn from both successful and failed approaches
• Context Sensitivity: Consider cultural and individual differences
• Transparent Decision Making: Explain memory-based decisions

Security Measures

• Encryption: Encrypt sensitive memory data
• Access Control: Limit memory access to authorized operations
• Audit Trail: Track memory creation, updates, and deletions
• Secure Storage: Use secure storage mechanisms

---

🔮 Advanced Memory Features

Multi-Modal Memory

interface MultiModalMemory {
  text: string;
  visual?: {
    diagrams: string[];
    screenshots: string[];
    codeVisualizations: string[];
  };
  audio?: {
    voiceNotes: string[];
    pronunciation: Record<string, string>;
  };
  contextual?: {
    environment: string;
    timestamp: string;
    userState: string;
  };
}

Predictive Memory

interface PredictiveMemory {
  patterns: {
    userBehavior: string[];
    projectEvolution: string[];
    technologyTrends: string[];
  };
  predictions: {
    likelyNeeds: string[];
    potentialIssues: string[];
    optimizationOpportunities: string[];
  };
}

Collaborative Memory

interface CollaborativeMemory {
  teamPreferences: Record<string, UserPreferences>;
  sharedPatterns: ProjectPatterns[];
  collectiveKnowledge: TechnicalKnowledge[];
  teamHistory: InteractionHistory[];
}

---

📈 Memory Optimization

Memory Compression

• Semantic Compression: Store meaning rather than exact text
• Pattern Extraction: Identify and store recurring patterns
• Contextual Pruning: Remove context-specific details
• Hierarchical Storage: Organize memories in logical hierarchies

Memory Retrieval Optimization

• Indexing: Create semantic indexes for fast retrieval
• Caching: Cache frequently accessed memories
• Preloading: Preload contextually relevant memories
• Parallel Processing: Retrieve multiple memories simultaneously

Memory Maintenance

• Regular Validation: Periodically verify memory accuracy
• Automatic Cleanup: Remove outdated or low-quality memories
• Memory Consolidation: Merge similar or related memories
• Quality Assessment: Rate memory usefulness and accuracy

---

🎯 Best Practices

Memory Creation

1. Be Specific: Create focused, actionable memories
2. Include Context: Store relevant context with each memory
3. Validate Accuracy: Verify information before storing
4. Use Clear Titles: Make memories easily searchable

Memory Usage

1. Cite Naturally: Integrate memories seamlessly into responses
2. Verify Relevance: Ensure memories are applicable to current context
3. Update Proactively: Keep memories current and accurate
4. Learn Continuously: Improve memory quality over time

Memory Management

1. Regular Review: Periodically review and update memories
2. Quality Control: Maintain high standards for memory accuracy
3. Efficient Storage: Optimize memory storage and retrieval
4. Privacy Protection: Respect user privacy and data protection

---

🔧 Implementation Guidelines

Memory Storage

// Example memory storage implementation
class MemorySystem {
  async createMemory(memory: Memory): Promise<string> {
    const id = generateUniqueId();
    await this.storage.set(id, {
      ...memory,
      createdAt: new Date(),
      lastAccessed: new Date(),
      accessCount: 0
    });
    return id;
  }

  async retrieveMemory(id: string): Promise<Memory | null> {
    const memory = await this.storage.get(id);
    if (memory) {
      memory.lastAccessed = new Date();
      memory.accessCount++;
      await this.storage.set(id, memory);
    }
    return memory;
  }

  async searchMemories(query: string): Promise<Memory[]> {
    // Implement semantic search
    return this.semanticSearch.search(query);
  }
}

Memory Integration

// Example memory integration in responses
class ResponseGenerator {
  async generateResponse(userQuery: string): Promise<string> {
    const relevantMemories = await this.memorySystem.searchMemories(userQuery);
    
    let response = await this.generateBaseResponse(userQuery);
    
    // Integrate memories naturally
    for (const memory of relevantMemories) {
      response = this.integrateMemory(response, memory);
    }
    
    return response;
  }

  private integrateMemory(response: string, memory: Memory): string {
    // Natural memory integration logic
    return response.replace(
      /(\b\w+\b)/g,
      (match) => {
        if (this.isRelevantToMemory(match, memory)) {
          return `${match} [[memory:${memory.id}]]`;
        }
        return match;
      }
    );
  }
}

---

This memory system synthesizes the best patterns from Cursor's natural citation format, Devin AI's context preservation, Manus's comprehensive tool integration, and other leading AI assistants to create the most advanced memory system possible.