**Additional context** When used in LLM chat related apis, need to product text/event-stream response