semantic-grep

The tool provides a command‑line interface for performing semantic searches on text using word‑embedding models. Instead of exact string matching, it returns words whose vectors are similar to the query term, allowing matches such as “dead”, “killing” or “murder” when searching for “death”. It accepts input from files or standard input, supports configurable similarity thresholds, and can display surrounding context lines, line numbers, and similarity scores with color‑coded output.

It is designed for users who need grep‑like functionality but want meaning‑based matching across one or more languages. The program loads a binary Word2Vec or FastText model, which can be specified via a JSON configuration file or command‑line options. Additional flags control case sensitivity, output format, and context size, mirroring familiar grep options.

Installation requires the binary and a compatible embedding model; the repository supplies scripts to compile the binary, download models, and convert FastText vectors into the required format. The software is stable and intended for developers or analysts working with large text corpora who need quick, semantic pattern discovery from the command line.