Code Sage

Provides high-performance semantic code search combining BM25 keyword search with vector embeddings and RRF reranking, using AST-based chunking to intelligently split code into semantic units across 60+ programming languages for natural language code exploration and analysis.

Code Sage is a high-performance semantic code search server written in Rust that combines BM25 keyword search with vector embeddings using RRF reranking for hybrid search capabilities. The implementation uses AST-based chunking via tree-sitter to intelligently split code into semantic units (functions, classes, methods) across 60+ programming languages and file types, with character-based fallback for unsupported formats. Built with embedded storage (USearch for vectors, Tantivy for full-text, Sled for metadata) requiring zero external dependencies, it supports multiple embedding providers including OpenAI, LM Studio, and Ollama, with automatic .gitignore respect and custom file extension support. The server exposes tools for code analysis, natural language search, index management, and status checking, making it valuable for code exploration, documentation generation, and development workflows requiring semantic understanding of large codebases.