This paper concentrates on the important techniques of design and implementation of Interface Agent (NanDa User Interface Agent, NDUIA) in DOLTRI-Agent (distance and open learning training resource information retrieval agent). This system extends the memory-based reasoning technique, using multiple memory models and multiple analysis models, and the authors can generate a specific interest's model for each user by analyzing his/her using experience. At the same time, through the combination of user interests model and using experience in special scene, interface agent can provide some active intelligent services, such as information guidance, preprocess of the searching results, automatic modification of classification information, intelligent and instant help, etc. In this way, the authors achieve collaboration between software and its users.