Presently, smart interconnected heterogeneous devices with built-in GPS, Wi-Fi, and Bluetooth capabilities are more affordable, resulting in numerous novel mobile applications. To adapt applications based on the environment they are executed in, we perform context management by proactively monitoring the (network-and location-based) context information available in such devices and applications. As this requires continuous monitoring, the common fixed data flow based technique for forwarding contexts from multiple context sensors is not energy efficient. In this paper, we propose the design and implementation of an adaptive context monitoring scheme. Alongside context awareness, we employ overlay network, agent, and policy theories. We utilize learning and personalization characteristics to make an optimized judgment of context information in an efficient and scalable fashion. This paper is complemented with protocol evaluations to validate scalability claims based on real logs.