The aim of the ID4CS system is to solve a design problem. The system to design will contain several optimization methods such as FSQP, derivative-free approaches, methods based on genetic algorithms, etc. and several disciplinary models at different precision levels of description. We propose to encapsulate each method and model in software agents, and to create one multi-agent per level. Thus, the architecture proposed is composed of several levels. Each level is an adaptive multi-agent system and matches with a given level of granularity of the models. One level is dedicated to the methods, another is the abstract level of disciplines, and another is a more specialized point of view about the disciplines.

• A module enabling the communication by messages passing between agents. The interaction within a level and inter-level interaction are not alike and differ by the content of the communication. The information exchanged between two agents of different levels requires the resolution of interoperability problems,
• Modules enabling the reasoning, learning and the agent behaviour control,
• A module representing the knowledge about what an agent has or gains, about its neighbours in the system,
• A module representing the knowledge related to a discipline.

The expert knowledge is composed of the knowledge about the discipline and some additional meta-information such as uncertainties, validation range, response time, etc. which can improve the behaviour of the agents to consequently improve the resolution process. This system will be tested on the preliminary aircraft design and on module engine design.The aim is to work in trying to provide a generic tool which can be applied to other domains such as: car design, helicopter design, ship design.