Replymessage unavailable
Mainly inference in normal application flows, not federated training. My first step is to classify the data and check the provider's retention and access terms, then keep sensitive material behind a private endpoint even if the model weights are public. Your gateway distinction is useful too.