This talk outlines Sandwood, an open-source, type safe, Java like probabilistic programming language, compiler, and runtime. Sandwood aims to make it easier to include Bayesian models in applications by providing a language that is familiar to developers and by taking advantage of the more complete view of the model available to compiled languages. Bayesian models written in Sandwood are compiled to Java classes which can be instantiated into objects where each object represents one instance of the model. Each of these instances can be configured to target different hardware and execution models and these configurations can be update during the lifetime of the object.