In this paper we propose an intelligent agent model for smart home environments. In ubiquitous computing environments like smart home, agents have to learn user's preferences in order to assist them. These preferences are represented by user profiles. Our proposed UT-AGENT uses case-based reasoning (CBR) to acquire knowledge about user actions that are worth recording to determine their preferences and Bayesian Network (BN) as an inference tool to model relationships between them. UT-AGENT maintains the status of every device present in the home and activates them as per user preference. It generates a sequence of expected user's query and simultaneously activates the devices with preset preference settings.