Context-aware applications require local access to current and relevant views of context information derived from global sensors. Existing approaches provide only limited support, because they either rely on a network broker service precluding open-ended searches, or they adopt a presence model which has scalability issues. To this end, we propose a fully distributed architecture employing context user-agents co-located with data-mining agents. These agents create and maintain local schemas using ranking of global context information based on context proximity. Continually evolving context information thus provides applications with current and relevant context views derived from global sensors. Furthermore, we present an evaluation model for assessing the effort required to present local applications with current and relevant contextual views. We show in a comparison with earlier work that the approach achieves the provisioning of evolving context information to applications within predictable time bounds, circumventing earlier limitations.