This paper presents an approach to automatically create a knowledge base for smart buildings and let it dynamically evolve based on the behaviors of the occupants inside. Such a knowledge base compromises all life-style experiences and intelligent buildings. Accordingly, management decisions and control can be performed. To achieve the goal, an online learning algorithm is used which is designed to catch knowledge and generate rules from environmental sensing data and actuators states. Rules are fed to the system as fuzzified events. Applying such a learning approach to maintain the knowledge base, abstracts the complicated coordination between human, infrastructure components and static structures such as building floorplan and architecture.