Wireless channel charting is a self-supervised machine learning approach designed to capture and leverage the statistical properties of wireless propagation. By applying dimensionality reduction to channel state information, channel charting establishes a low-dimensional representation of the channel state, which is akin to a pseudo-location of the users within the propagation environment. Conversely to multipoint channel charting, which considers efficient ways of fusing the features acquired by multiple access points, we introduce in this work the multi-site channel charting problem whereby multiple base stations generate their own channel charts, while their coverage areas partially overlap. We introduce several methods to align the latent spaces at each base station into a common latent space, effectively performing the fusion of multiple channel charts. We benchmark the proposed approaches and compare their performance in terms of the classical dimensionality reduction metrics using measured data, as well as their degree of distributedness.