As far as we know we can use input sentence get an array which contains the input results. So different sentences get the different array based on our sentence input engine.
And also there are a lot of decisions represented by the output sentence, we need transform that to the execute information by our sentence output engine and send them to agent.