Résumé : The computer science and more precisely the networks, integrate more and more our life. In order to evaluate the impact, the ergonomics and the performances of new technologies, researchers and industry people use, regularly, simulation or emulation techniques. When these simulations put together several simulators, we deal with co-simulation. The co-simulations and their current architectures do not provide sufficient flexibility and modularity. Because of their important centralization, they are difficult to use on a larger scale. The emergence or the improvement of technologies like multi-agents systems or web services allow to propose networks of independent applications, supervised by an artificial intelligence. Complex systems could be built by assembling basic modules driven by agents. This work uses results obtained in the field of co-simulations to validate choices made to define our co-simulation platform. Among different possible technological solutions we have chosen a combination of them that answer to our scope statement. We extend this solution for co-simulations to a cooperation of applications. Moreover, in order to fulfill the need of developing a chain of image processing and to prove the feasibility of such architecture, we have proposed a platform that allows to heterogenous applications to collaborate. The chain of image processing, that we study, rebuilds a 3D environment using images provided by a camera. The created data are shared with other customer applications that increased this way their knowledge. This chain uses applications of computer vision able to process the informations very quickly, allowing thus a (almost) real time reactivity.