monkesearch/monkeSearch

Implements semantic file search through dual architectures: an LLM-based approach that converts natural language queries directly into macOS Spotlight predicates (no indexing required), and a vector database backend using sentence transformers with temporal regex parsing for cross-platform support on Linux/Windows. Extracts three semantic components from queries—file type, temporal context, and keywords—then grounds results against actual file metadata (creation/modification dates, MIME types, content) rather than free-form similarity alone.

174 stars.