A short tour on first launch shows where everything is: personas, attachments, the mic and new chats. It plays once, and you can skip it. The example prompts on the empty chat screen, and the Chef, Tutor, Writer and Analyst personas, now appear in your language.

New: Qwen3, an optional open model. Choose it in Settings and download it once (about 1.5 GB from Apple), and Onmind can answer with Qwen3 instead of Apple Intelligence. It runs on your iPhone's Neural Engine, and what you type still never leaves your device. Fewer "I can't help with that" replies to everyday questions. Ask Siri or Shortcuts to search Onmind, and it opens your past conversations filtered to what you asked for. In Japan, you can set the side button to open Onmind's voice chat. Read aloud, the new design and this update's new features are now in all twelve languages. Onmind 2.3 requires iOS 27.

Faster, steadier answers. Onmind no longer gets stuck calling the same tool again and again, so its slowest answers now finish in seconds instead of minutes. More accurate. Tools are used only when a question needs them, so everyday questions get a direct answer, and a word problem with several steps is worked out in one calculation. Unit conversions put the answer first and use the US cup and gallon. Skills no longer start just because a word like "today" or "split" appears in a message. Clearer when it can't help: on iOS 27 a request Onmind won't answer gets a plain reply instead of a system message, and a conversation that runs long recovers on its own. Read any answer aloud with the speaker button. Onmind uses the most natural voice on your iPhone, supports your Personal Voice, and shows how to get a free Premium voice if you only have the basic ones. Fixes a crash when you tap the microphone, and a rare crash when your Apple Watch gets an answer from your iPhone.

Skills: Onmind now follows a written procedure for jobs you do often. Split a bill, plan your week, read a receipt, build a packing list, or set up a study plan. It tells you which skill it used, and you can turn any of them off in Settings. Now in twelve languages: English, French, German, Italian, Spanish, Portuguese (Brazil), Japanese, Korean, Chinese (Simplified and Traditional), Dutch and Swedish. Long answers appear on the Lock Screen and in the Dynamic Island, with a Stop button. A Control Center button starts hands-free voice chat, and it works from the Action button too. Follow-up questions are faster: a conversation now keeps one on-device session instead of starting over each turn. Your Apple Watch answers math, unit conversions and dates on its own when your iPhone is out of reach. Better code help: Onmind now checks edge cases before telling you code is fine, and names the exact input that breaks it. Fixes: links, tables and nested formatting now read correctly in read-aloud and exports; decimal commas work in calculations; relative dates like "in 7 days" work in the calendar and date tools.

• Apple Intelligence Foundation Models: on-device conversational AI running directly on Apple Silicon and the Neural Engine, with no server round-trip. • Five Tailored Personas: switch between General Assistant, Code Helper, Creative Writer, Data & Document Analyst, and Study Tutor from the toolbar. • Real-Time Token Streaming: responses render token by token with live throughput telemetry. • Markdown Rendering: formatted replies with headings, lists, emphasis, and code blocks. • Export Your Conversation: copy any chat out as clean Markdown. • Press Return to send, Shift-Return for a new line. • 100% Offline: works in Airplane Mode. No account creation, no network requests, no telemetry.

• Dictation now runs entirely on your iPhone — your voice is never sent to a server. • Conversations name themselves with smart, on-device titles. • New Apple Watch complication shows your last answer at a glance. • Replies render headings and lists cleanly. • A "Try Again" button on any reply that didn't go through. • Voice mode fixes: it no longer talks over itself, and your music stays playing. • Fixed a crash when clearing chat history, and tapping Stop now keeps the partial answer.