When you send a screenshot, add a photo of yourself, or we write your words, that image or text goes to our server and on to Google’s Gemini AI, which reads it for us. Your three lines are written from the text it read, her chat or her profile, and its one-line notes on her photos, with anything you type under a result: by OpenAI’s GPT if you agreed to this policy in Gatzby 1.2.1 or later, otherwise by Gemini. OpenAI gets that text through Vercel’s AI Gateway and never gets an image. A photo goes with one-line notes on the other photos in your six; your words are written from those notes and, when you ask for one line again, from your current lines. When you ask What’s my move?, the result’s screenshots go to Gemini once more, and the text it reads from them, her chat or her profile and its notes on her photos, goes with the result’s three lines to Anthropic’s Claude AI, which writes the read on her; Claude never gets an image. Each request also carries a random id made on this phone, so we can count free uses. No name, no account, no contacts, no location.