How is this different from using GPT api on Azure? I thought that allowed you to keep you data corpus/documents private as well, ie not get sent to their servers for training