Ubiquitous computing environments rely on the extensive use of networked sensors which acquire real-time information about people and their activities. This information can then be collected and processed in order to support our daily life activities in terms of comfort, safety and efficiency. For instance, in the specific context of assisted living, this makes it possible to improve the autonomy, independence and self-esteem of elderly people and of people with different kinds of disabilities (visual-auditory-mobility handicaps). One approach that is being explored in the MIDAS project [7] has been to model different viewpoints of the problem such as the person competencies and abilities, the function and behavior of the services and applications that are available in his/her environment and the situation at hand when the person is engaged in his/her activities. These models are based on domain ontologies that capture the semantic content of sensors data. Such viewpoints are denoted “contexts”. In this paper, we present this approach and show how we have used this approach to develop a prototype of an intelligent assistant, that we call CAAS (Context Aware ASsistant). The CAAS detects and reacts to critical situations by planning, scheduling and conducting the appropriate course of actions.