Comment by 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.