This work is part of a project dedicated to the development of a cardiac arrhythmia monitoring system. The Calicot system is composed of two distinct parts: a temporal abstraction part, dedicated to the acquisition, the processing and the analysis of the signal, and a medical diagnosis part which computes a diagnosis from the data transmitted by the temporal abstraction. To increase the system performances, we decided to add to our system a pilot whose goal is to choose, according to the context, the type of information which is needed and the best way to get it from the signal. This pilot has three levels: the arrhythmia recognition pilot is in charge of determining the level of detail at which information is needed according to the current diagnosis hypotheses; the temporal abstraction pilot is in charge of (de)activating the abstraction tasks according to the signal characteristics and the recognition requirements; the signal processing pilot is in charge of deciding which algorithm is the most adapted to perform a selected abstraction task and tuning it according to the context and to the other pilots requirements. We present the new architecture of our system in which the pilot plays a central role. Then, the first results, mainly got at a signal processing level, are analyzed and we show the feasibility and the interest of such an approach.