In order to provide the best possible on-line services according to the needs of individuals and companies, the concept of Quality of Service (QoS) has become more and more important. Formerly provided on high-speed networks such as ATM, the QoS concept has become part of the Internet thanks to new resource reservation protocols like RSVP. To allow users to make the most out of the limited resources at their disposal, it is necessary that networked-based multimedia applications adapt their quality according to users' choices and available resources. As part of ACR Labs' researches on a Multi-Agent QoS Management System, this project aimed at conceiving and developing an agent in charge of the acquisition of user's preferences. It guides the adaptation to hardware conditions according to those preferences. Three modules compose this agent: the traditional Graphical User Interface, the Learning Module with some simple Artificial Intelligence functions, and the Communication Module also in charge of the translation of user's choices to application settings. The Java implementation, simulating teleconference and video-on-demand services, only considers a local-network-based video application. Only the first tests could be held so far. Although the agent's accessibility must be improved before using it in real conditions, it seems to be a good experimental tool. Experiments and improvements are thus proposed at the end of the report.