Multivariate time series classification is a challenging task where black box models achieve high performances. However, in real-world applications, interpretability is crucial for helping users understand the decision-making process of an algorithm, not just its performance. In this paper, we present a multi-agent ensemble learning approach for time series classification suited for online learning. Our approach relies on the organization of agents in the feature space at each time steps. We demonstrate that our approach achieves performances comparable to state-of-the-art methods. Finally, we highlight its explainability and interpretability properties as a white-box model.