In spoken dialogue applications dialogue management has conventionally been realized with a single monolithic dialogue manager implementing a comprehensive dialogue control model. We present a highly distributed system structure that enables the integration of different dialogue control approaches to handle spoken dialogues. With this structure it is possible to integrate multiple dialogue control models into a single working application in a flexible fashion. The main principle is that all the processing is distributed into many compact agents that focus on single tasks. The agents process the information independently and share all the information via shared information storage. We present the principles that enable such distribution. In addition, a multilingual example application, AthosMail, is presented.