Sem is a semantic code search tool using the all-MiniLM-L6-v2 model for embeddings and FAISS as the vector database.

Embeddings are generated and stored locally in ~.sem/index/. None of your code leaves your computer.

This codebase is made up of a VsCode extension written in Typescript called semsearch, as well as a code search tool called semvector built using Python.

Semsearch provides an interface through which you can index and query your codebase.

Semvector creates a chunked dictionary representation of a codebase, which is then passed off to a local model to generate embeddings. These embeddings can then be queried against to semantically search your codebase.

Sem is not currently available in the VsCode extension store as, well, it needs some work. I mean don't get me wrong, it works, but it can be a lot better.