Help with understanding ML-Agent's communication process

Hi, I'd like to understand how ML-Agents interact with Unity, and possibly, write an interface which can send the relevant state action and reward from another simulator to the training process of ML-Agents. Where should I start with the Unity - ML-Agents communication framework? How does Unity send the states to ML-Agents?

 

have you checked into the Unity.MLAgents.CommunicatorObjects folder?