Somax2 is a multi-agent interactive system, based on machine-listening, machine learning and generative units, performing live machine co-improvisation with musicians. The system is trained on any musical materials chosen by the user (Audio or MIDI), effectively constructing a generative model, a corpus, from which it draws its musical “knowledge”. Listening to and adapting to a musician, that activates dynamic profiles, Somax2 provides stylistically coherent improvisation in real-time. As part of the REACH project, defining co-creativity as distributed agency between human and machine, Somax2 is indeed a good example of a distributed co-creative system, where multimodal cross-feedback affects collective behavior and increases player’s engagement through emotion and motivation.