This research proposes a new model for analyzing and correcting non-schedulable partitions in real-time multiprocessor systems, specifically in the context of fault tolerance in distributed networks. The need for such a model arises from current techniques for correcting non-schedulable partitions that must be revised and repartitioning all tasks across processors. The proposed model is based on intelligent agents and implemented using the JADE platform. The model consists of (1) a supervisor agent in the first layer that distributes tasks and manages system correction when a non-schedulable partition is detected; and (2) a second layer composed of partition agents that analyze schedulability, request corrections, and negotiate with the supervisor for additional tasks to correct the entire system. The effectiveness of the proposed model is demonstrated through a case study. Quantitative analysis shows that the proposed model improves fault tolerance in distributed systems and has the potential for further enhancement by adding communicative tasks, heterogeneous processors, and other improvements.