The first-launch wizard. The chosen options are highlighted in gold. At the bottom — an example of how Jarvis will greet you.

1 · Language

Русский or English. Everything switches at once: speech recognition, the voice, commands and help. You can also switch by voice: switch to Russian, then переключись на английский.

2 · Who will use it

A man or a woman. This decides which forms of address the next question offers.

3 · How to address you

For a man: sir (default), Mr + surname or by first name. For a woman: ma’am (default), miss — for a young woman, Mrs + surname — for a married one, or by first name. Just like Jarvis with Tony Stark.

4 · Who will do the thinking

Built-in — on this Mac. The Gemma model runs on the computer: no internet, nothing leaves the machine. Takes about 13 GB of memory.

External — ChatGPT, Claude… After launch the settings open by themselves at the “Brain” section, where you enter a key. How to get one and where to put it — section 05.

What happens next

• Intro music — if enabled in the settings: Jarvis’s own theme or your file. It fades as soon as Jarvis starts speaking.
• The greeting out loud and as a line under the clock: “Good afternoon, sir!..”. The line disappears once the phrase is spoken.
• The briefing: how much is unread in Mail and messengers (by the badge numbers in the Dock, the contents are not read) and what the weather is.
• The microphone turns on by itself. Otherwise answering to his name would make no sense. You can see the state on the “Hearing” indicator under the clock.

The program window

Jarvis opens full screen. It is an ordinary macOS program: you can minimise it, switch to another and come back. The window position is remembered.

Quit the program like any other (⌘Q) or by voice: Good night, Jarvis. He will say goodbye and exit. All farewell and wake phrases are in section 11.

Graphics and look

Graphics quality, Pick graphics automatically, Frames per second, Render scale and Save the GPU while the brain thinks — section 04.1. Layout for the “Grid” view: arc, grid, spiral. Main screen: orbit, callouts, classic. Folders: around the sphere or as a grid.

Gestures and files

Gesture sensitivity · show hidden files · content previews · mirror the camera.

Brain — who thinks

The most important row: here you choose who answers your requests — the built-in Gemma or an external AI (ChatGPT, Claude, Gemini, DeepSeek, Alice, Claude Code). Next to it — the model, the key field and the “Check” button. Details in section 05. Below: Fine-tuned brain (the adapter for the built-in model), Notice interesting things in what is open, Turn the movie down when I speak.

Parental control and the guide

Parental control — the child’s age and the parent’s password (section 04.2). User guide — four buttons: open this book right inside Jarvis, open the copy on the Desktop, save the PDF anywhere, or open it on the website.

Intro music

Jarvis’s theme, your own file, or off. For your own file — from which second, how long to play, the “Clip” button to listen, music volume.

Name and waking

Answer to names — a comma-separated list, the first one is the main. Answer out loud to the name — “Yes, sir?”. Wake by claps: enable, restart, start at login. Fall asleep after silence — how many seconds after answering Jarvis waits for a command without the name (section 11).

Speech

Address and the “Change…” button (the same wizard). Language of commands and answers. Assistant voice: Karen (main), Karen-2, Serena, Vivian — for Russian and English. Manner of speech: calm, neutral, fast. Liveliness of intonation, pace, pauses between phrases, answer volume and the “Check” button — a test phrase. Audio devices — where the sound goes and where we listen from.

At the bottom of the panel

“Voice commands” — the built-in help. “Download the hand model” — if the hand model is missing.

Settings → graphics. The auto-pick box is ticked, so the fields below show what was chosen and are locked. In the frame — what Jarvis found: the GPU, the monitor’s refresh rate and the result.

The rows in order

Settings → “Parental control”. The age, the on/off box, the parent’s password and the “Apply” button. Below — what is on right now.

How to turn it on

1. Open the settings (⌘,) and find the row “Parental control — a child at the computer”.
2. Choose the age: 6+, 12+ or 16+, and tick the box.
3. The first time the field is called “Make up a parent password”: enter a password of at least 4 characters.
4. Press “Apply”. The line “Parental control is on: 12+” appears.

What it looks like

• In the built-in browser a closed site does not open: instead there is a notice “closed by parental control”.
• To a spoken request to find or open such a thing Jarvis replies: “This is closed by parental control” — and names the topic.
• Other browsers Jarvis will not launch on request, so the control cannot be bypassed through him.
• Movie and file names are checked by explicit words; Jarvis does not know the age rating of the movie itself.

Change the age or turn it off

Choose another age or untick the box, enter the parent’s password and press “Apply”. Without the password the changes are not saved: “The parent password did not match”. The password is stored on the Mac only in encrypted form.

Connecting an external brain — step by step

1. Get a key on the website of the chosen model (where exactly — the table on the next page). The key is a long string of letters and digits. Copy it whole.
2. Open the settings (⌘,) and in the row “Brain — who thinks” pick the model from the list. Until there is a key, the name is marked “needs a key”, and the built-in one keeps thinking.
3. In the “Model” row pick a variant: “smarter” or “faster”. The default one is fine to start with.
4. Paste the key into the “Key …” field and press “Save”. The field clears and shows “saved ✓”.
5. Press “Check”. Jarvis asks the model a short question and shows whether it answered and in how many seconds. If all is well, he says: “Working with Claude, sir”.
6. Press “Done”. From now on the external model thinks. The built-in one is unloaded from memory and frees about 13 GB.

The “Brain” rows in the settings after choosing Claude: the list of brains and the “Check” button, the model, the key field with the “Save” button and a hint where to get the key.

What opens inside

• Photos — at full size.
• Video and music — the built-in player with seeking, pause and full screen (section 14).
• PDF — the built-in viewer.
• Text — notes, .md, .txt.
• Letters, web pages, WhatsApp and Telegram — in windows too (section 15).

Window commands

The name rule

A command starts with the name: Jarvis, open downloads. Without the name Jarvis does not react — so a movie soundtrack and other people’s conversations never become commands. The name alone, without a command, gets a short “Yes, sir?”, and he keeps listening.

After a command is done Jarvis listens for another 10 seconds without the name: “Done, sir. Any further instructions?”. Small things — volume, pause, minimise — he does silently. How he wakes and falls asleep — section 11.

Interrupting

While Jarvis is speaking, say Jarvis, stop, enough, not that or let me clarify, I need… — he falls silent and listens. Just to cut the answer short: quiet.

If several were found

Jarvis shows a numbered list and asks. Answer the first, number three, the last, name the file (february), say all or cancel. You can also click the options with the mouse.

Numbers and names

Names can be said in your language even if the file is written differently: “donald” finds donald_face.png. Numbers are understood as words: “version fifteen” → v15. Formats — by ear: “pdf”, “word”, “excel”, “jpeg”.

Jarvis is closed — claps

1. Clap your hands twice — sharply, one after another. A short silence is needed before the first clap.
2. A quiet signal sounds — the claps were heard, Jarvis is waiting for the phrase.
3. Within six seconds say the wake phrase (the list on the right). The long variant triggers on its first words already.
4. Jarvis opens: the boot screen, intro music, the greeting and the briefing — what came in and what the weather is.

The claps are heard by a small background program, “Jarvis — Wake” (next page). Its microphone only watches the loudness, and the phrase is recognised on the Mac; nothing is recorded or sent anywhere. While Jarvis is open, it lets go of the microphone.

Phrases that wake him after the claps

Any name from the settings (“Answer to names”) works. “On Friday” is a weekday, not a call. To check whether the program hears the claps, look at the icon in the menu bar: it says “Waiting for ‘Wake up, sleepyhead’”.

Jarvis is open — the name

The name alone — “Yes, sir?”, after that the command can be said without the name. The name together with a command — straight to work. How many seconds after answering he waits for a command is set by “Fall asleep after silence” (45 s by default); then he waits for the name again. This is the “sleep” of hearing: the microphone works, but reacts only to the name.

Find and act at once

Where to look: “on the computer”, “on the mac”, “everywhere” — across the whole home via Spotlight; without these words — in the current folder. Type: movie, video, photo, document, music, archive, folder, program, spreadsheet, presentation. Format: “in pdf format”, “with the docx extension”.

Open

Programs

Move and copy

Change

Delete — to the Trash only

Selection and order

View and details

Windows and the screen

Sound and volume

Movies and music

Weather

Mail and documents

Internet

Assistant

The built-in help. Opens with the phrase what can you do or the “Voice commands” button in the settings.

How it works

• Searches the Documents, Desktop and Downloads folders through macOS search — builds no index of its own, uses no memory, answers in a fraction of a second.
• Reads PDF and scans (recognises the text), Word, RTF, Excel, PowerPoint, web pages and plain text.
• In a long document finds the most relevant passages — even from the middle of a fifty-page report.
• Answers strictly by the text and names the source document.

Reads what is open

Opened a letter, a PDF or an article in the browser — ask what is this about? or what does the letter ask for?. If you turn on “Notice interesting things in what is open” in the settings, Jarvis points out the main thing himself whenever you open something — no more than once in 40 seconds and never interrupting you.

“Remember that…”

Say Jarvis, remember: car inspection on October 14 at 10 am.

• A note appears in the folder Documents › Jarvis › Notes — an ordinary text file with the date in its name. It is visible in Finder; old notes are never overwritten.
• A week later ask when is the car inspection? — Jarvis answers from the note.
• Notes can be opened and edited like any file.

What the player can do

• Any sound. AC3, E-AC3, DTS and TrueHD — what usually lives in MKV — are decoded inside Jarvis and stay in sync with the picture.
• Tracks. Russian by default, otherwise the main one. Switch by voice: switch to the original, bring back the Russian dub, put on the Vizgunov translation.
• Seeking. skip 20 seconds, back (10 seconds), to the 45th minute, from the start. Pause, continue, full screen — by voice or with the player buttons.
• Titles out loud — by meaning. “Distant Stars”, not “Distant.Stars.2024.1080p.BluRay.x264.mkv”. On screen — the full name.

Volume and ducking

• turn it up · quieter · volume 40% · mute the sound · unmute.
• Turn the movie down when I speak — the setting is on by default: as soon as you start speaking the sound goes down, and comes back after the answer.
• Sound on Bluetooth headphones? Jarvis listens to you through their microphone.

If it did not play

A rare format that could not be read — Jarvis offers to open it in IINA or VLC: “‘…’ does not play here. Play it in IINA?”. Answer “yes”. To open in an external player right away: open the movie in IINA.

The “Mail” widget. The latest letters from the Mail app; in the right corner — how many are unread.

Mail

Jarvis reads letters from the Mail app — read only, he never sends or deletes anything. Ask what came in the mail? or what is in the letter from accounting? — he reads it and points out the main thing: “They ask you to sign the act by Friday”. A click on a letter in the widget opens it in a window; if Mail is closed — iCloud Mail opens inside Jarvis.

The briefing after the greeting

Unread items are counted by the badge numbers in the Dock: Mail, Messages, Telegram, WhatsApp, Viber, Slack and other running programs. The contents are not read and no permissions are needed.

Tips

• The mouse always wins. Move the mouse — control returns to it; let it rest — the hand works again.
• Sensitivity in the settings: how small a hand movement crosses the whole screen.
• Mirror the camera — if the movements feel “backwards”.
• Even light on the hand and a plain background help a lot.

The iPhone camera

An iPhone near the Mac appears in the list as an ordinary camera (Continuity Camera) and is offered first. If the iPhone is asleep or far away — pick the built-in camera in the drop-down list on the gesture panel.

The hand model

Recognition of 21 hand points runs on the Mac, without a network. If the model file is missing, a simplified tracker works — the cursor and grabbing without individual fingers; the full version comes back with the “Download the hand model” button in the settings (about 7 MB, once).