gabriele06-local/localmind

LocalMind is a local semantic search engine written in Rust that indexes your text, PDF, and Word documents and lets you search them by meaning rather than exact keywords. When you type a query like "how to fix memory leaks" it finds documents about that topic even if they never use those exact words. The AI model behind this is a compact BERT variant called all-MiniLM-L6-v2, which runs entirely on your CPU with no internet connection, no cloud service, and no API key required. The tool has a terminal-based user interface where you type a query and see results with similarity scores in about 55 to 65 milliseconds. Results are navigable with arrow keys and show which file matched. A background polling loop watches your document folder and automatically re-indexes any file that changes, so the search index stays current without you having to trigger a rebuild manually. Under the hood, the vector index is stored in a single binary file you can copy or back up like any other file, rather than requiring a separate database server. The file format is designed for fast sequential reading: each vector sits at a known offset in a flat array, and the CPU can stream through them efficiently. The similarity calculation uses SIMD instructions, a technique that processes eight numbers at once using dedicated CPU hardware, which is why queries return in tens of milliseconds even on modest machines. The search also combines traditional keyword matching with the vector search and merges the two result lists using a method that gives you the precision of exact-term matching alongside the meaning-awareness of the AI model. The release binary is about 8 MB including the model loader and the full search pipeline, making it straightforward to copy to another machine. It is designed for a personal document collection of up to a few thousand files. The README does not specify a license.