This article presents a computational system which uses concepts of an artificial intelligence technique known as intelligent agents aggregated to human-computer interaction concepts. Its main objective is to help the improvement of a determined web interface interaction and usability -applied to the teaching at distance. The architecture was developed over a LGPL (Lesser General Public License) platform that includes intelligent agents, two bases of knowledge with information obtained through heuristic evaluations and, afterwards transformed in rules. In order to verify the co-operation of the computational system two types of evaluations were performed: heuristic, to verify the help referring to the usability improvement and cognitive trajectory to verify the user's learning to manipulate the interface. The result was the comparison of time and the verifications of some questions in the web interface screens that show the user's performance with or without the utilization of the developed computational system.