In this work, an ultra-local model-based predefined-time assist-as-needed controller (UPTAC) is presented for a upper limb patient-exoskeleton system (ULPES) under input and performance constraints. The designed UPTAC has a dual-loop control architecture. The outer impedance sub-control loop includes task performance function and impedance controller, which dynamically adjust the speed of rehabilitation exercises and obtain the desired assistive torque. Next, under the ultra-local model-based control framework, an inner torque sub-control loop was designed by combining barrier Lyapunov function and performance constraints, thereby achieving predefined time convergence and reducing the complexity of controller design. Finally, the effectiveness of UPTAC was verified through simulation.