Peer-to-peer (P2P) video streaming systems enable video data exchange between peers by reducing the overload of the servers while utilising network resources. Multi-agent systems (MASs), including many autonomous and intelligent software agents working on behalf of video peers, may also provide a feasible infrastructure for the implementation of P2P video streaming systems. Within this context, research on the use of software agents in video streaming systems has recently emerged. In this study, the authors discuss the development of an agent-based P2P video streaming system. Software engineering of the MAS with covering various aspects such as agent internals and interactions in the system is described. Performance evaluation of the proposed agent-based P2P system comparing with the popular in-use streaming application is also reported. Results show that a well-defined behaviour of a parent selection software agent can improve the continuity index in P2P video streaming systems. Therefore the users in the system can watch video in a better quality and lower end-to-end delay comparing with the currently used video streaming applications. They also examine that peer welcoming (traffic localisation) behaviour of the agents decreases the internal Internet service provider traffic significantly.