Integrated modeling approaches (multi-simulation, multimodeling, etc.) have proven challenging in practice. The first challenge deals with the technical aspects of coupling different computational or mathematical components. The second challenge lies in the alignment of the semantics of these components so that their integration does make sense, which is particularly critical in pluridisciplinary models. A number of approaches have been proposed in the last 20 years but none of them is really suitable to our context . We propose in this thesis an alternate approach, called co-modeling, which borrows concepts and tools from agent-based modeling, agent-oriented software engineering and multimodel ecologies. Simply speaking, a co-model can be defined as a multi-agent system of models and datasets. Each model or dataset is represented by one or several agents interacting with one another within the context of a larger representation of their — potentially dynamic — environment. The proposed approach does not aim at providing a general solution to the two challenges above, but at providing a framework in which modelers can easily implement their solution or test different coupling solutions. The proposed approach is fully implemented within the GAMA agent-based modeling platform. Its advantages are shown in terms of flexibility, composability and reusability in a number of case studies. The first case study is the dynamic coupling of equation-based and agent-based models to obtain “switching” models dynamically. The second one is the design of a complex integrated model where three formalisms and four modeling approaches have been successfully coupled.