In a peer-to-peer system, providing a service consumes peers resources and leads to the apparition of selfish peers keeping their resources for their own use. Selfish peers decrease or suspend their service offer while requesting services from others. This behavior overwhelms peers offering services and degrades the system performance. Incentive mechanisms reduce or eliminate selfishness and encourage collaboration. In this paper, we propose a peers' classification based on their selfishness or contribution level according to which peers are sanctioned or rewarded. With this objective, we introduce a new parameter called the activity level of a peer. Moreover, we describe the architecture of an incentive mechanism that uses the previous classification for a monitoring component called the Activity Level Controller. We show how to couple the different components for an incentive mechanism and give the pseudo-code of each building block of the incentive mechanism. Simulations results show that the success rate of the active peers is much higher than that of the inactive peers and blacklisted peers. Also, the messages overhead is lower than in a system without incentive mechanism.