Assistant teams extend the present view on personal assistance software. Instead of a single assistant, a team devotes an ensemble of agents to the private use of each user. It can provide a new variety of functionalities relying on the composition of individual services for the owner benefit. In this paper, we present a model of assistant teams based on the exploitation of extended agent interactions. In addition to the usual direct interactions among individuals, we introduce the concept of 'overhearing' to let agent react to indirect events. We describe architecture of such systems, address the concern of modeling their interaction infrastructures with a construction methodology, and detail an example scenario.