We aim at designing a virtual organisation of exchanges between teachers and learners in an open distant teaching situation based on artificial intelligence tools, such as multi-agent paradigm or Bayesian networks. In addition, we propose to provide teachers with appropriate tools to manage on-line teaching and to assess student's knowledge acquirements. Traditional systems put teachers or students at the heart of the process. We intend to focus our approach more on the interaction that exists among them. When talking about interaction, we mean all the possible exchanges (Student to Teacher, Class to Teacher...) that could be defined, as well as by their intrinsic nature, their frequency or their contents. The use of a shared scheduler will make possible all global synchronization. To deal with these interactions, we propose to introduce different kinds of software agents that reproduce human behaviours. We intend to design a student modelling system, which uses a standard behaviour of student provided either by an appropriate learning or by an a priori description, and adapted to each specific situation. This profiling step should allow taking into account any specificity and could be viewed as an estimation of the student's state. From this model we plan to diagnose any abnormal situation and, if necessary, to trigger alarms in a step of notification. This diagnostic could be viewed as a detection of lakes (lacks) or peaks (strengths) in the student knowledge landscape and in its psychological state. As the aim of the tool is to prevent any abnormal situation before it occurs, different levels of alarms may be triggered. In a first level, the student may be warned that there is something going wrong and may receive a predictive help. In a second step, teachers may receive alerts for a need of interaction with the student.