This paper presents multi-agent intrusion detection system (IDS) operating based on asynchronous data streams arriving from multiple heterogeneous sources. IDS is composed of many structured specialized classifiers each trained to detect attacks of a fixed class. Alerts produced by classifiers having the same specialization and operating in different feature representation spaces are correlated at the upper layer. The top-layer classifier solves intrusion detection task: it combines heterogeneous alerts produced at a previous layer. The paper describes IDS architecture and emphasizes the reusability of some solutions used in the developed IDS. The implemented multi-agent IDS is capable to detect three classes of attacks, DoS, Probe and U2R using input traffic.