The downside is that it's slower to encode documents during indexing compared with an external embedding end-point unless you use a GPU instance for Typesense -- that will speed up auto embedding generation significantly.