Modern neural activity recordings, gathered across multiple trials, conditions and subjects, present an exciting opportunity to explore brain-wide dynamics. Current computational models, however, often fail to leverage the richness of such data, either by providing uninterpretable representations (e.g., deep networks) or oversimplifying brain dynamics. Here, instead of regarding asynchronous neural recordings that lack alignment in neural identity as a limitation, we leverage these diverse views into the brain to learn a unified model of neural dynamics. We assume that brain activity is driven by multiple hidden global circuits that capture core interactions between neural ensembles—functional groups of neurons—such that the time-varying decomposition of these circuits defines how the ensembles' interactions evolve over time non-stationarily. With our new model, termed CREIMBO (Cross-Regional Ensemble Interactions in Multi-view Brain Observations) we identify the hidden composition of per-session neural ensembles and model the ensemble dynamics on a low-dimensional manifold spanned by a sparse time-varying composition of the circuits. CREIMBO disentangles overlapping temporal neural processes while preserving interpretability due to the use of a shared underlying sub-circuit basis. We demonstrate CREIMBO's ability to recover true components in synthetic data, and using mouse whole-brain recordings, we show its ability to discover dynamical interactions that capture meaningful variables.