Weblogs are being used by enormous number of users. Finding a weblog that matches the interests of readers, is one of the challenges that still persists. Recommendation systems might be a solution. I introduce a new class of weblogs and describes the design of a multi-agent recommender system which utilizes elements such as blogroll links between blogs, semantic similarities and visit statistics. Recommendations are made by modeling the links between weblogs based on the trust between them and inferencing.