AgentDB — семантический векторный поиск для RAG

ai-tooling

agentdb-vector-search is a Claude Code skill that implements semantic vector search using AgentDB's high-performance vector database, designed for building RAG pipelines, semantic search engines, and intelligent knowledge bases. It uses HNSW indexing, binary and scalar quantization, and delivers sub-millisecond queries under 100 µs — benchmarked at 150x to 12,500x faster than traditional solutions. The skill supports hybrid search combining vector similarity with metadata filters, Maximal Marginal Relevance for diverse result sets, and batch inserts; preset configurations cover collections of under 10K, up to 100K, and over 1M vectors. Accessible via CLI (`npx agentdb@latest`) or a TypeScript API, it also runs as an MCP server inside Claude Code, exposing agentdb_query, agentdb_store, and agentdb_stats tools. A good fit for developers adding meaning-based retrieval over documents using OpenAI embeddings or custom embedding models.