Nowadays most “LLM” endpoints include some sort of server side harness as well, and I’d bet more than one model involved, so it’s really just agents all the way down