Abstract. In order to cope with load management in power grid systems, this paper presents a hybrid multi-agent framework. This framework integrates the advantages of both centralized and decentralized architectures to achieve both accurate decisions and quick response, and avoid the single point of failure as well. The development of various agents and the behaviors of each agent in the framework are described. Moreover, an example is also introduced, which demonstrates the interaction among agents when a fault happens in a power grid system. The contribution of this paper is to combine local intelligence with global coordination in multi-agent system design to satisfy the challenging requirements in a power grid system.