[timrogers]

That would be the normal pattern. But you could certainly stop after the LLM picks the tool and provides the arguments, and not present the result back to the model.