At present, distributed intrusion detection systems are taking on more and more characteristic of distributing and intelligence. Multi-agent system (MAS), which shows its intelligence by the interaction of a group of intelligent agents, is a popular tool for building novel intrusion detection systems (IDSs). The BDI (Belief-Desire- Intention) model is a well studied architecture of agency for intelligent agents situated in complex and dynamic environments. Traditional MAS-based distributed IDSs commonly do their detection work within a framework that employs the method of distributed data collection and hierarchical data analysis. This is a simple and strict method, but it restricts the characteristic of distributing, intelligence and real-time response badly. In this paper, we present a dynamically-created hierarchical framework, and then bring forward the basic algorithm system to support it. We introduce the opponent BDI model to our agents, and realize the detection based on multi-agent cooperation, which effectively improves the traditional distributed intrusion detection.