When NOT to Use This Skill

• Local-only operations with no vector search needs
• Simple key-value storage without semantic similarity
• Real-time streaming data without persistence requirements
• Operations that do not require embedding-based retrieval

Success Criteria

• Vector search query latency: <10ms for 99th percentile
• Embedding generation: <100ms per document
• Index build time: <1s per 1000 vectors
• Recall@10: >0.95 for similar documents
• Database connection success rate: >99.9%
• Memory footprint: <2GB for 1M vectors with quantization

Edge Cases & Error Handling

• Rate Limits: AgentDB local instances have no rate limits; cloud deployments may vary
• Connection Failures: Implement retry logic with exponential backoff (max 3 retries)
• Index Corruption: Maintain backup indices; rebuild from source if corrupted
• Memory Overflow: Use quantization (4-bit, 8-bit) to reduce memory by 4-32x
• Stale Embeddings: Implement TTL-based refresh for dynamic content
• Dimension Mismatch: Validate embedding dimensions (384 for sentence-transformers) before insertion

Guardrails & Safety

• NEVER expose database connection strings in logs or error messages
• ALWAYS validate vector dimensions before insertion
• ALWAYS sanitize metadata to prevent injection attacks
• NEVER store PII in vector metadata without encryption
• ALWAYS implement access control for multi-tenant deployments
• ALWAYS validate search results before returning to users

Evidence-Based Validation

• Verify database health: Check connection status and index integrity
• Validate search quality: Measure recall/precision on test queries
• Monitor performance: Track query latency, throughput, and memory usage
• Test failure recovery: Simulate connection drops and index corruption
• Benchmark improvements: Compare against baseline metrics (e.g., 150x speedup claim)

AgentDB Memory Patterns

What This Skill Does

Use this skill to implement memory management patterns for AI agents using AgentDB's persistent storage and ReasoningBank integration. Apply these patterns to enable agents to remember conversations, learn from interactions, and maintain context across sessions. Deploy triple-layer retention (24h/7d/30d+) for optimal memory organization.

Performance: 150x-12,500x faster than traditional solutions with 100% backward compatibility.

Prerequisites

Install Node.js 18+ and AgentDB v1.0.7+. Ensure you have AgentDB via agentic-flow or standalone. Review agent architecture patterns before implementing memory systems.

Quick Start with CLI

Initialize AgentDB

Run these commands to set up your AgentDB instance with memory patterns:

# Initialize vector database
npx agentdb@latest init ./agents.db

# Or with custom dimensions
npx agentdb@latest init ./agents.db --dimension 768

# Use preset configurations
npx agentdb@latest init ./agents.db --preset large

# In-memory database for testing
npx agentdb@latest init ./memory.db --in-memory

Start MCP Server for Claude Code

# Start MCP server (integrates with Claude Code)
npx agentdb@latest mcp

# Add to Claude Code (one-time setup)
claude mcp add agentdb npx agentdb@latest mcp

Create Learning Plugin

# Interactive plugin wizard
npx agentdb@latest create-plugin

# Use template directly
npx agentdb@latest create-plugin -t decision-transformer -n my-agent

# Available templates:
# - decision-transformer (sequence modeling RL)
# - q-learning (value-based learning)
# - sarsa (on-policy TD learning)
# - actor-critic (policy gradient)
# - curiosity-driven (exploration-based)

Quick Start with API

import { createAgentDBAdapter } from 'agentic-flow/reasoningbank';

// Initialize with default configuration
const adapter = await createAgentDBAdapter({
  dbPath: '.agentdb/reasoningbank.db',
  enableLearning: true,      // Enable learning plugins
  enableReasoning: true,      // Enable reasoning agents
  quantizationType: 'scalar', // binary | scalar | product | none
  cacheSize: 1000,            // In-memory cache
});

// Store interaction memory
const patternId = await adapter.insertPattern({
  id: '',
  type: 'pattern',
  domain: 'conversation',
  pattern_data: JSON.stringify({
    embedding: await computeEmbedding('What is the capital of France?'),
    pattern: {
      user: 'What is the capital of France?',
      assistant: 'The capital of France is Paris.',
      timestamp: Date.now()
    }
  }),
  confidence: 0.95,
  usage_count: 1,
  success_count: 1,
  created_at: Date.now(),
  last_used: Date.now(),
});

// Retrieve context with reasoning
const context = await adapter.retrieveWithReasoning(queryEmbedding, {
  domain: 'conversation',
  k: 10,
  useMMR: true,              // Maximal Marginal Relevance
  synthesizeContext: true,    // Generate rich context
});

Memory Patterns

Apply these memory patterns to organize agent memory across multiple time horizons.

1. Session Memory

Implement session-based memory to track conversation history:

class SessionMemory {
  async storeMessage(role: string, content: string) {
    return await db.storeMemory({
      sessionId: this.sessionId,
      role,
      content,
      timestamp: Date.now()
    });
  }

  async getSessionHistory(limit = 20) {
    return await db.query({
      filters: { sessionId: this.sessionId },
      orderBy: 'timestamp',
      limit
    });
  }
}

2. Long-Term Memory

// Store important facts
await db.storeFact({
  category: 'user_preference',
  key: 'language',
  value: 'English',
  confidence: 1.0,
  source: 'explicit'
});

// Retrieve facts
const prefs = await db.getFacts({
  category: 'user_preference'
});

3. Pattern Learning

// Learn from successful interactions
await db.storePattern({
  trigger: 'user_asks_time',
  response: 'provide_formatted_time',
  success: true,
  context: { timezone: 'UTC' }
});

// Apply learned patterns
const pattern = await db.matchPattern(currentContext);

Advanced Patterns

Hierarchical Memory

// Organize memory in hierarchy
await memory.organize({
  immediate: recentMessages,    // Last 10 messages
  shortTerm: sessionContext,    // Current session
  longTerm: importantFacts,     // Persistent facts
  semantic: embeddedKnowledge   // Vector search
});

Memory Consolidation

// Periodically consolidate memories
await memory.consolidate({
  strategy: 'importance',       // Keep important memories
  maxSize: 10000,              // Size limit
  minScore: 0.5                // Relevance threshold
});

CLI Operations

Query Database

# Query with vector embedding
npx agentdb@latest query ./agents.db "[0.1,0.2,0.3,...]"

# Top-k results
npx agentdb@latest query ./agents.db "[0.1,0.2,0.3]" -k 10

# With similarity threshold
npx agentdb@latest query ./agents.db "0.1 0.2 0.3" -t 0.75

# JSON output
npx agentdb@latest query ./agents.db "[...]" -f json

Import/Export Data

# Export vectors to file
npx agentdb@latest export ./agents.db ./backup.json

# Import vectors from file
npx agentdb@latest import ./backup.json

# Get database statistics
npx agentdb@latest stats ./agents.db

Performance Benchmarks

# Run performance benchmarks
npx agentdb@latest benchmark

# Results show:
# - Pattern Search: 150x faster (100µs vs 15ms)
# - Batch Insert: 500x faster (2ms vs 1s)
# - Large-scale Query: 12,500x faster (8ms vs 100s)

Integration with ReasoningBank

import { createAgentDBAdapter, migrateToAgentDB } from 'agentic-flow/reasoningbank';

// Migrate from legacy ReasoningBank
const result = await migrateToAgentDB(
  '.swarm/memory.db',           // Source (legacy)
  '.agentdb/reasoningbank.db'   // Destination (AgentDB)
);

console.log(`✅ Migrated ${result.patternsMigrated} patterns`);

// Train learning model
const adapter = await createAgentDBAdapter({
  enableLearning: true,
});

await adapter.train({
  epochs: 50,
  batchSize: 32,
});

// Get optimal strategy with reasoning
const result = await adapter.retrieveWithReasoning(queryEmbedding, {
  domain: 'task-planning',
  synthesizeContext: true,
  optimizeMemory: true,
});

Learning Plugins

Available Algorithms (9 Total)

1. Decision Transformer - Sequence modeling RL (recommended)
2. Q-Learning - Value-based learning
3. SARSA - On-policy TD learning
4. Actor-Critic - Policy gradient with baseline
5. Active Learning - Query selection
6. Adversarial Training - Robustness
7. Curriculum Learning - Progressive difficulty
8. Federated Learning - Distributed learning
9. Multi-task Learning - Transfer learning

List and Manage Plugins

# List available plugins
npx agentdb@latest list-plugins

# List plugin templates
npx agentdb@latest list-templates

# Get plugin info
npx agentdb@latest plugin-info <name>

Reasoning Agents (4 Modules)

1. PatternMatcher - Find similar patterns with HNSW indexing
2. ContextSynthesizer - Generate rich context from multiple sources
3. MemoryOptimizer - Consolidate similar patterns, prune low-quality
4. ExperienceCurator - Quality-based experience filtering

Best Practices

1. Enable quantization: Use scalar/binary for 4-32x memory reduction
2. Use caching: 1000 pattern cache for <1ms retrieval
3. Batch operations: 500x faster than individual inserts
4. Train regularly: Update learning models with new experiences
5. Enable reasoning: Automatic context synthesis and optimization
6. Monitor metrics: Use stats command to track performance

Troubleshooting

Issue: Memory growing too large

# Check database size
npx agentdb@latest stats ./agents.db

# Enable quantization
# Use 'binary' (32x smaller) or 'scalar' (4x smaller)

Issue: Slow search performance

# Enable HNSW indexing and caching
# Results: <100µs search time

Issue: Migration from legacy ReasoningBank

# Automatic migration with validation
npx agentdb@latest migrate --source .swarm/memory.db

Performance Characteristics

• Vector Search: <100µs (HNSW indexing)
• Pattern Retrieval: <1ms (with cache)
• Batch Insert: 2ms for 100 patterns
• Memory Efficiency: 4-32x reduction with quantization
• Backward Compatibility: 100% compatible with ReasoningBank API

Learn More

• GitHub: https://github.com/ruvnet/agentic-flow/tree/main/packages/agentdb
• Documentation: node_modules/agentic-flow/docs/AGENTDB_INTEGRATION.md
• MCP Integration: npx agentdb@latest mcp for Claude Code
• Website: https://agentdb.ruv.io

---

Core Principles

AgentDB Memory Patterns operates on 3 fundamental principles:

Principle 1: Triple-Layer Memory Retention

Agent memory organizes across three time horizons: immediate (10 messages), short-term (session context), and long-term (persistent facts), mirroring human memory architecture.

In practice:

• Store recent messages in immediate memory for context window management
• Persist session-level context (goals, preferences) in short-term memory
• Extract important facts to long-term memory for cross-session retrieval
• Consolidate memories periodically to prevent unbounded growth

Principle 2: Pattern Learning via Reinforcement

Agents improve through experience by storing interaction patterns and training on successful/failed attempts using 9 RL algorithms.

In practice:

• Record trigger-response patterns with success indicators and context
• Train learning plugins (Decision Transformer, Q-Learning, Actor-Critic) on pattern corpus
• Retrieve proven patterns via vector similarity for new situations
• Update pattern confidence scores based on real-world performance

Principle 3: Reasoning-Augmented Retrieval

Context synthesis combines multiple memory sources with reasoning agents (PatternMatcher, ContextSynthesizer, MemoryOptimizer) for rich contextual understanding.

In practice:

• Use retrieveWithReasoning() instead of raw vector search
• Enable synthesizeContext: true to generate coherent context from fragments
• Apply Maximal Marginal Relevance (MMR) to reduce redundancy in results
• Leverage ExperienceCurator to filter low-quality memories

---

Common Anti-Patterns

Anti-Pattern	Problem	Solution
Flat Memory Structure	Storing all memories at same priority causes retrieval noise and unbounded growth	Implement hierarchical memory with immediate/short-term/long-term layers
No Memory Consolidation	Keeping all interactions forever exhausts storage and degrades retrieval quality	Run periodic consolidation with importance-based pruning and size limits
Raw Vector Search	Retrieving embeddings without reasoning misses contextual relationships	Use retrieveWithReasoning() with synthesizeContext: true for richer results
Ignoring Learning Loops	Never training on stored patterns wastes collected experience	Train learning plugins regularly with adapter.train() after accumulating experiences
Storing Everything	Recording trivial interactions pollutes memory with low-value data	Filter by importance/relevance; store only interactions with confidence >0.5

---

Conclusion

The AgentDB Memory Patterns skill enables stateful AI agents through persistent cross-session memory with triple-layer retention and pattern learning capabilities. By combining HNSW-accelerated retrieval, 9 RL algorithms, and reasoning-augmented context synthesis, it provides 150x-12,500x faster memory operations than traditional solutions.

Use this skill when building conversational agents, creating intelligent assistants that learn from interactions, or implementing systems requiring context persistence across sessions. The pattern library excels at chat systems, task-planning agents, and recommendation systems where historical interactions inform future decisions.

Key takeaways: Organize memory hierarchically across time horizons, train learning plugins on accumulated patterns, and leverage reasoning agents for context synthesis. The 100% backward compatibility with ReasoningBank and seamless Claude Code integration via MCP makes AgentDB Memory Patterns the production-ready choice for stateful agent development.