got it, but a semantic search would require setting up a vector database. it cant be done using the API. so yeah, latency + setting up the vector db will be additional overheads.