a minimal vector storage in numpy.
build your index of documents and query it using cosine, euclidean or manhattan similarity.
python3 test.py
so, does it scale? no.
but if your application doesn't need more than a couple hundred thousand entries, why make things more complicated than they need to be?