This paper focuses on decentralized personalized search engines. It is composed of three parts. Firstly, we formulate the problem and we propose a graph-based measure of quality of a document given a user and a word. This quality gives a formal measure to test the personalized relevance of answers provided by search engines. Secondly, we present a decentralized system namely Maay which aims to provide personalized results to its users. Users can share, search and retrieve documents from others. Nodes locally learn others? profile from previous interactions and use feedbacks to raise words characterizing a document. By querying preferentially nodes with similar profile and by ranking documents according to the requester profile, nodes may find documents which are relevant for them. Finally, we propose a framework for experimental studies of such system. Then, we use it to show that the main principles of Maay could benefits to users.