||In many joint military operations, different allies have their own machine learning models trained for various analytical and decision-making tasks. These models are generally trained using localized training data often characterized by different modalities of input signals. As a result, these different models may not be able to utilize or perform well on the combination of all the available signal data to make critical decisions during the operational stage. In order to share the knowledge embedded in different models trained independently in each individual domain, we propose the concept of policy- based ensembles. This approach allows one to combine hetero- geneous models from different domains into an ensemble, whose operations are controlled by policies specifying which subset of the models ought to be used for an operation. These policies are based on the metadata containing information about the training datasets as well as the individual model characteristics shared by the individual domains. This demonstration aims to show the benefits of our policy-based ensembles over the naive ensemble approach.