Abstract: In this paper, a new control method, i.e. agent-control, is presented. In this method, a control algorithm is encapsulated in a control entity which can move and reuse in the network. Agent-control includes two modes:control-on-demand and a combination of simple locally control and complicated remotely control. This control method is modeled by UML(Unified Modeling Language). Due to the limitation of UML in representing agent, the UML is extended by virtual classdiagram. Part of code of an example for agent-control is given.