Energy consumption of computers in office buildings remains an issue of concern for companies. For example, for a university campus, they might represent between 50% and 80% of all the power consumed. Besides that, computer usage behavior sometimes makes the application of energy saving policies a difficult task, as many users prefer not to be annoyed by waiting for computers to wake up. We present in this paper a novel software architecture to enable dynamic and static appliance of energy saving policies in office computers while observing and taking into account user behavior. The key is taking advantage of idle periods for the computers and the localization of the user to determine when to turn Off or On the PC and maximize the savings. We ran a study at our campus where we found that idle periods represented between 30% and 50% of the On periods. We simulated the savings for our policies and could potentially achieve savings between 32% and 38% of current power consumption numbers. With even more tuning, we believe savings could be higher and similar techniques could be used for other devices in buildings.