Zay-M3/NaturalSQL

Extracts SQL database schemas and vectorizes them using configurable backends (Chroma or SQLite) with local or Gemini embeddings, enabling semantic retrieval of relevant tables for LLM context. Supports PostgreSQL, MySQL, SQL Server, and SQLite with automatic in-memory caching of embedding models (~10-15ms subsequent searches) and vector storage reuse to avoid redundant indexing. Provides lightweight schema-to-vector pipeline specifically designed for RAG workflows without heavy framework dependencies.