In a distributed multi-agent based software environment, the traditional monolithic user model ceases to exist and is replaced by user model fragments, developed by the various software agents populating the environment. These model fragments have been developed in a variety of specific contexts to help achieve various goals. User models are thus fragmented, relativized, local, and often quite shallow. They are inherently inconsistent with one another and reflect not only characteristics of the users, but also certain social relationships among them. With the arising proliferation of models and data, the user modelling problem transforms into retrieval and integration of the available user model fragments "just in time" by a particular computational agent to the breadth and depth needed for a specific purpose. In this paper we explore the implications of distributed user models, drawing examples from I-Help, a collaborative system for peer help.