Assistant (talk to Zeus)
Ask Zeus by voice and hear the answer. Open it with the round microphone button at the bottom right of the desktop (beside the dock), the microphone button on lite's home screen, or Super+J. What you said and the answers stay in the Zeus app's chat — open the Zeus app and the same conversation goes on.
What it does
- Optional search keys: Zeus › Settings › Tools › Personal search keys opens Vault. Add entries named
TavilyorExa. Search also works without keys; keys improve reliability. Zeus never reads their values.
- Talk: hold the microphone button while you speak and let go. A single tap listens until you pause (about 0.65 s of quiet).
- Writing it down: the sound goes to our server's speech recognition (Whisper) and only text comes back. The sound is not kept anywhere — the server turns it into text in memory and drops it, and the device deletes its temporary file as soon as it is sent. Logs hold only sizes and times.
- Two answers: a spoken question gets two answers — one for the ear and one for the screen. The spoken one sounds like a colleague explaining out loud ("first…", "and…", "by the way…"), with no lists, parentheses, screen labels like "Schedule:" or dates like 2026-11-11 (it says "Wednesday, November 11"). Its length follows the question — short for simple things, complete when an explanation is needed. The screen shows the organised answer (headings, lists, numbers); the voice never reads the screen text. A briefing sounds like a morning radio show (an opening, schedule, weather and news joined by transitions, a closing); news like an anchor, one or two sentences an item. Weather and news are fetched when asked, and the voice does not talk about when they were fetched — if a lookup fails it says so once ("I can't get the weather right now"). If you pause and go on talking before the answer starts, your words are joined and asked again as one question (no answer per fragment). While a tool runs (web search, weather, news…) the screen says "Looking it up…" and the voice says "Let me look." once.
- Asking about now: for today's news this device asks the news sources itself (Google News, Naver news and others; not through our server) and keeps articles from the last three days, the stories most outlets carry first: four when spoken, up to eight in Studio, each with its outlet and report time. A topic with nothing in three days gets the last two weeks, and it says so. The last line names where it looked. Today's weather and the current time are checked when asked. If a lookup fails, the Assistant says why. With Web search off in Settings › Tools, it does not search for news either.
- Barge-in (speaker mode): the microphone stays open while the answer is read. Start talking and the reading stops at once and your new words are taken. Its own voice from the speaker does not stop it: the level of what it plays is the reference. Phone browsers also use echo cancellation. In the Mac app (Hear and Read on the Mac) it takes two steps: 0.12 s of speech pauses the reading at once (within 0.3 s), more than 0.48 s ends it and takes your words; if it goes quiet for 0.3 s before that (a short 「yes」, 「mm」) the reading goes on from where it paused.
- Hands-free: when on, it listens again after the answer. When off, it listens only when you press.
- Follow-up chips: 2–4 next questions under the answer; tap one or press
1–4. Chips are never read aloud. On opening: "What can I do for you?" and chips of things it does; after 10 s of quiet only the chips change. - App buttons: app links (
zeus://…) in the answer are buttons. - App demos ("Show me"): ask "show me a Calendar demo", or press an app's ⚡ Zeus and then its first chip "Show me": Zeus opens the app and does a short task with sample data at a pace you can follow — the pointer moves and presses, and the strip at the bottom shows the step. The device checks every step and writes it into the chat as a numbered line (a step that did not happen gets ✗ and the reason), the screen is recorded and attached. At the end: what the app is for, and chips to ask next. Your files are not touched (even the history and settings files an app writes on its own are put back). Esc, Stop, any key or press stops it. The recording goes to Movies › 제우스 데모 when the home folder has room (three are kept, the oldest goes); otherwise to the Mac folder (Mac › NanumOS › 제우스 데모), and then an Allow card showing the place and the size asks every time. Cancelled, or neither place: only the last screen stays in the chat (gone after a restart). Desktop only for now.
Screen
- Top: "Assistant", the state (Listening… · Writing down… · Thinking… · Looking it up… · Speaking), a waveform, End.
- Middle: the conversation (your words on the right in blue, answers on the left). The last bubble is the answer.
- Bottom: the chips, the big microphone button, the "Read answers aloud" and "Hands-free" switches.
- While it listens the top bar shows a red microphone pill.
- Desktop: a panel at the top centre;
Escor a click outside closes it. Lite: the whole phone screen becomes the conversation.
Shortcuts
| Key | Does |
|---|---|
Super+J (hold) | talk while held |
Super+J (tap) | open and listen until a pause |
Space · Enter | talk again / end what you say |
1–4 | ask that chip |
Esc | stop the reading; again to close |
| any other key | stop the reading |
Settings
read 1|0— read answers aloud (on by default here)handsfree 1|0— hands-free (on by default)autosend 1|0— the Zeus app's microphone sends at once (off by default: the words land in the box to edit and send)voice 0..7— who reads: Minseo (default), Minseo bright, surprised, worried, annoyed, Junho, Sua, Jihun — the voices of the Nanum voice pack. English sentences use the Mac's voice for now (Mac app): the pack has no English voice yetmacvoice 1|0— in the Mac app, read with the Mac's own voice (off by default: Minseo). Who is reading now is in the check's voice row ("read with the Nanum voice 「민서」")
Microphone
- Lite (phone): the browser asks once (https only). Echo cancellation, noise suppression and auto gain are on. The screen stays on while the conversation is open (Wake Lock, where the browser has it).
- The Mac window (NanumOS in a QEMU window): the Mac's microphone. A small helper on the Mac that starts and stops with the window (
ZeusOsCam, the same one as the camera) passes the sound to NanumOS — the Mac's microphone is on only while the Assistant listens (the orange dot at the top of the Mac's screen). If the Mac has not allowed the microphone yet, the Assistant says "The Mac is not letting the microphone through" and exactly what to do: on the Mac, System Settings › Privacy & Security › Microphone, switch on "Terminal" (the app that started the window). Then speak again. NanumOS.app (the VZ window) has no microphone yet.
Not yet
- Approval cards ([Approve] [Edit] [Cancel]) for deleting, sending or changing settings are drawn here once Zeus sends that card; for now it only says "Please confirm on the screen".
- Follow-up chips come from Zeus when it sends them; otherwise built-in ones.
- Backchannels ("yeah", "uh-huh") also stop the reading (김비서 tells them apart with a transcript; not yet here).
- No on-device speech recognition yet — without the server it says "Could not reach the voice server".
- On NanumOS the speaker echo is told apart by level only (a headset avoids it).
Version 1.0.0
NanumOS 0.1.8349 · 2026-11-16