This is my app, the mode is Agent, no tool is called, and the model is a privately deployed LLM When I issue a query, the Agent's thinking process is streamed out, but the time result of the request ...