This research work presents an adaptive and embodied virtual tutor. Our proposed tutor is able to adapt the execution of a peda-gogical scenario according to the estimated student's level of knowledge. To achieve such a goal, we rely on Mascaret, a meta-model for knowledge representation in a virtual environment and on an inference of the student's memory content. This inference permits the tutor to adapt the execution of the pedagogical scenario and to choose an individualized assistance according to the evolution of the student.