Hmm I see thanks. But my plan is to have the custom LLM backend built in with workflow management (eg with langgraph) to do rag for example and other controlled flows before responding back. Intuitively, that way workflow is built in my custom LLM. Am I missing anything important?