Many e-learning portals have appeared on the Web to promote distance learning. However the amount of information available on these web sites increases more and more and users do not know where to find the “good” information. It is become necessary to build additional tools to help learner in their learning. Nowadays few works consider the idea to integrate intelligent tutors for such environments in ever-changing evolution and social-constructivism approaches show the importance of interactions between actors (learners, tutors) in the learning process.

To answer these problems, this paper presents CAMELEON, a generic model of embodied conversational agent (ECA) devoted to e-learning portals. This agent is always in interactions with the learner in order to improve his learning task. The agent is only the visible part of an underground multi-agent system, in which several agents interact according to the user actions, the web site evolution and the pedagogical scenarios. All these interactions allow to filter all the available information in order to adapt the ECA behavior to learners.