In order to be used in various contexts of training, a desirable property for a virtual environment for training, especially a collaborative one, is adaptability. In this paper, we identify two levels of adaptability the environment should offer. The first level concerns the parametering of the components of the application, especially the scenario and the actors' behavior. The second level appears during the execution of the application where the actors should be able to adapt the scenario xecution to the context. The models proposed in this paper fully satisfy those requirements. The scenario, the virtual world, the actors and the training session can be parametrized. Moreover, an action selection mechanism enables the actors to dynamically choose an action to perform depending on both the current context and the past actions. Finally we validate our models by presenting an illustrative scenario within the GVT environment.