Designed as a platform to enable rich contextual data collection in real homes, BoxLab uses a broad array of wireless sensing devices to study responsive applications situated in natural home settings. BoxLab has been deployed in homes around the Boston area, and has generated a dataset containing over 10,000 hours of sensor data to be used as training libraries for computational activity recognition and other applications of artificial intelligence. BoxLab also enables rapid deployment of context-triggered applications that allow systems to respond to occupant activities in real time.