The Free Energy Principle (FEP) and its extension, Active Inference (AIF), present a unique approach to understanding self-organization in natural systems. These frameworks propose that agents use internal generative models to predict observations from unknown external processes, continuously updating their perceptive and control states to minimize prediction errors. While this unifying principle offers profound insights into agent-environment interactions, implementing it in practical scenarios poses significant challenges. Researchers require fine-grained control over agent-environment communication protocols, particularly when simulating proprioceptive feedback or multi-agent systems. Current solutions from reinforcement learning and control theory, such as Gymnasium, need more flexibility for these complex simulations.