Advances in agent technologies have fueled interest towards combining agent-based techniques with tools from data mining. Furthermore, the advent of the ubiquitous computing paradigm has introduced a novel applications area, which offers new possibilities for further deepening this bond. Ubiquitous computing solutions are often designed for mobile environments, which poses additional requirements on practical solutions. Providing a distributed architecture enables the distribution of workload, while at the same time allowing the on-line processing of data. Such an architecture ensures the pervasiveness of privacy and security aspects in the overall design. In this paper we present an architecture for distributed data preprocessing in ubiquitous environments, which supports the full distribution of processing tasks and enables the encapsulation of privacy and security mechanisms within every component. As our second contribution we discuss more thoroughly the usefulness of the intelligent agent paradigm for context-aware systems.