Ever-increasing quantities of personal data are generated by individuals, knowingly or unconsciouly, actively or passively (e.g., bank transactions, geolocations, posts on web forums, physiological measures captured by wearable sensors). Most of the time, this wealth of information is stored, managed, and valorized in isolated systems owned by private companies or organizations. Personal information management systems (PIMS) propose a groundbreaking counterpoint to this trend. They essentially aim at providing to any interested individual the technical means to recollect , manage, integrate, and valorize his/her own data through a dedicated system that he/she owns and controls. In this vision paper, we consider personal preferences as first-class citizens data structures. We define and motivate the threefold preference elicitation problem in PIMS-elicitation from local personal data, elicitation from group preferences, and elicitation from user interactions. We also identify hard and diverse challenges to tackle (e.g., small data, context acquisition , small-scale recommendation, low computing resources, data privacy) and propose promising research directions. Overall, we hope that this paper uncovers an exciting and fruitful research track.