Personalized services such as news recommendations are becoming an integral part of our digital lives. The problem is that they extract a steep cost in terms of privacy. The service providers collect and analyze user's personal data to provide the service, but can infer sensitive information about the user in the process. In this work we ask the question "How can we provide personalized news recommendation without sharing sensitive data with the provider?" We propose a local private intelligence assistance framework (PrIA), which collects user data and builds a profile about the user and provides recommendations, all on the user's personal device. It decouples aggregation and personalization: it uses the existing aggregation services on the cloud to obtain candidate articles but makes the personalized recommendations locally. Our proof-of-concept implementation and small scale user study shows the feasibility of a local news recommendation system. In building a private profile, PrIA avoids sharing sensitive information with the cloud-based recommendation service. However, the trade-off is that unlike cloud-based services, PrIA cannot leverage collective knowledge from large number of users. We quantify this trade-off by comparing PrIA with Google's cloud-based recommendation service. We find that the average precision of PrIA's recommendation is only 14% lower than that of Google's service. Rather than choose between privacy or personalization, this result motivates further study of systems that can provide both with acceptable trade-offs.