The World Wide Web (WWW) has become a vital service on the Internet. People are using the service day by day to get specific information or services they want. However, the traffic for the WWW service is increasing rapidly, and now all of the Internet backbone networks are congested by the WWW traffic. In order to improve this situation, several techniques to reduce the traffic have been developed: WWW client cache method, caching proxy servers and WWW hierarchical cache system. Furthermore, since the WWW cache system can be used for reducing the WWW latency, several active caching or "prefetching" method has been implemented. Our group designed and implemented the prefetching cache system called “Wcol” (WWW Collector) based on the idea of interactive prefetch scheme. In this paper, we show several performance results from its trial use at NAIST, Nara, Japan. Furthermore, on the AI3 network which is a research testbed network using satellite links covering several Asian countries, we have adopted a caching system with prefetching to improve both cache hit rate and usage of the bandwidth on the AB links. On top of caching and prefetching system, we add an agent to plan prefetching strategies based on observations of the traffic and the users' behavior. In this paper, we also show the design of this adaptive WWW cache system.