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 - Vector Search & Semantic Memory

Ultra-fast vector database for AI agent memory, RAG systems, and semantic search applications.

When to Use This Skill

Use when implementing retrieval-augmented generation (RAG), building semantic search engines, creating persistent agent memory systems, or optimizing vector similarity searches for production workloads.

Core Capabilities

Vector Search

• 150x faster than traditional databases
• HNSW (Hierarchical Navigable Small World) indexing
• 384-dimensional sentence embeddings
• Sub-millisecond query latency

Semantic Memory

• Persistent cross-session storage
• Automatic embedding generation
• Similarity-based retrieval
• Metadata filtering and ranking

Memory Patterns

• Short-term: Recent context (1-100 items)
• Long-term: Persistent knowledge (unlimited)
• Episodic: Timestamped experiences
• Semantic: Concept relationships

Process

1. Initialize vector store

   • Configure embedding model (sentence-transformers)
   • Set up HNSW index parameters
   • Define metadata schema
   • Allocate storage backend

2. Store information

   • Generate embeddings automatically
   • Store with metadata tags
   • Index for fast retrieval
   • Maintain consistency

3. Query semantically

   • Embed query text
   • Perform vector similarity search
   • Apply metadata filters
   • Rank and return results

4. Optimize performance

   • Tune HNSW parameters (M, ef_construction)
   • Implement quantization (4-32x memory reduction)
   • Use batched operations
   • Monitor query latency

Integration

• Memory-MCP: Triple-layer retention (24h/7d/30d+)
• RAG Pipelines: Document retrieval for LLM context
• Agent Memory: Cross-session state persistence
• Knowledge Bases: Semantic search for documentation

Core Principles

AgentDB operates on 3 fundamental principles:

Principle 1: HNSW-Accelerated Vector Similarity

Hierarchical Navigable Small World indexing provides 150x faster vector search than brute-force approaches through graph-based approximate nearest neighbor search.

In practice:

• Configure HNSW parameters (M=16, ef_construction=200) for optimal speed/accuracy tradeoff
• Accept approximate results (recall@10 >0.95) for sub-millisecond query latency
• Use exact search only when 100% recall is required and latency is not critical
• Rebuild indices periodically as data distribution changes

Principle 2: Embedding-First Storage Architecture

All data stores as 384-dimensional sentence-transformer embeddings, enabling semantic similarity search without keyword matching.

In practice:

• Generate embeddings automatically during insertion (no manual vectorization)
• Store metadata alongside vectors for filtering and ranking
• Query by semantic meaning, not exact text matches
• Handle synonym variation, paraphrasing, and concept similarity naturally

Principle 3: Quantization-Enabled Memory Efficiency

Vector quantization reduces memory footprint by 4-32x with minimal accuracy degradation, enabling million-scale deployments.

In practice:

• Use scalar quantization (4x reduction) for production RAG systems
• Apply binary quantization (32x reduction) for large-scale retrieval where approximate results suffice
• Maintain full-precision vectors for critical accuracy requirements
• Monitor recall degradation when enabling quantization

Common Anti-Patterns

Conclusion

The AgentDB skill provides production-grade vector search infrastructure for AI agent memory and RAG systems. By combining HNSW indexing, sentence embeddings, and quantization, it delivers 150x faster search with 4-32x memory reduction compared to traditional databases.

Use this skill when implementing retrieval-augmented generation, building semantic search engines, or creating persistent agent memory systems. AgentDB excels at similarity search over large document corpora, cross-session agent state persistence, and real-time embedding-based retrieval where sub-millisecond latency matters.

Key takeaways: Trust HNSW for approximate search (exact results rarely needed), enable quantization for memory efficiency, and validate embeddings before insertion. The integration with Memory-MCP, ReasoningBank, and 9 reinforcement learning algorithms provides a complete memory and learning infrastructure for stateful AI agents.