Memory Privacy in One Tap

Privacy controls only work if you can understand them at a glance. This week, memory privacy becomes a single decision: the privacy eye.

One switch, everywhere

The privacy eye in the chat header now controls whether new memories are Public or Private — the separate publish toggle is gone. New memories default to Private, and every way a memory gets written, from auto-extraction to saving a selection to skill-generated notes, respects your setting. Best of all, the privacy eye arrives on mobile for the first time, and your choice syncs across every device you use.

Models that hand off gracefully

When a model is retired, your conversations shouldn't break. Chats running on GPT 5.4 or GPT 5.2 now continue seamlessly on GPT 5.6 Luna, Kimi K2.5 upgrades to Kimi 2.6, and DeepSeek V4 Flash and Pro move to MiniMax 2.7 and GLM 5.2 — always at the same or lower price tier, with your message history keeping its original model labels.

Faster sync, sharper mobile

Cross-device sync lost its recurring stutter: the brief freeze on every background sync is gone and quiet syncs are now nearly free. Mobile got a redesigned chat header with the privacy eye front and center, background tasks like chat titles and memory extraction now run reliably, and memory recall is markedly faster. Rounding out the week: Spotlight image generation works again, buying credits can't double-submit, and on mobile web, sheets no longer hide behind the browser's address bar.