You say "embedding metadata in the custom LLM response if supported" is this like metadata added to the payload sent to an llm like openai for example