In this paper we present the new development of a semi-autonomous response module for the NOISA system. NOISA is an interactive music system that predicts performer’s engagement levels, learns from the performer, decides what to do and does it at the right moment. As an improvement for the above, we implemented real-time adaptive features that respond to a detailed monitoring of the performer’s engagement and to overall sonic space, while evaluating the impact of its actions. Through these new features, the response module produces meaningful and non-intrusive counter actions, attempting to deepen and maintain the performer’s engagement in musical interaction. In a formative study we compared our designed response module against a random control system of events, in which the former performed consistently better than the latter.