AgentDB — оптимизация векторного поиска

ai-tooling

agentdb-optimization is a Claude Code skill that optimizes AgentDB vector database performance through quantization, HNSW indexing, and caching strategies. It supports four quantization modes — binary (32× memory reduction, ~2–5% accuracy loss), scalar (4×), product (8–16×), and full precision — letting developers balance memory footprint against search accuracy depending on deployment scale. HNSW indexing is enabled automatically, delivering 150× faster searches (100 µs vs. 15 ms linear scan), while batch inserts process 100 vectors in 2 ms instead of 1 second. An in-memory LRU cache reduces repeated retrieval latency to under 1 ms. The skill targets Node.js 18+ projects using AgentDB v1.0.7+ via the agentic-flow package and is suited for applications scaling to millions of vectors.