This work defines the Interstellar as a recurrent neural architecture search problem for the short-term and long-term information along the relational paths, and proposes to search for recurrent architecture as the Interstellar for different KG tasks.