Pick a brain. Cloud is far smarter. On-device models download once and work offline. Chats and memory stay on your phone either way.
Enter your passcode.
Cloud models run on your own Cloudflare account, free up to a daily limit. On-device: your messages never leave the phone.
Things it remembers about you, saved only on this phone. Say "remember that…" in chat, or add one here.
When on, questions about news, weather, scores, prices or anything recent are looked up online first. Only the search words go out, through your own Cloudflare server. With an on-device brain, your messages themselves never leave the phone.
Natural voices sound human and run on your Cloudflare (the reply text is sent there to be spoken; they share the free daily limit, then fall back to the iPhone voice). iPhone voices stay on the phone; Premium ones (Settings › Accessibility › Spoken Content › Voices) sound best.
When there's no internet, Ben's AI switches to the on-device brain automatically. It needs Qwen downloaded once (about 1.3 GB) on Wi-Fi.