The 3D galaxy needs its graphics library from the internet, and it did not load. The orbit view is shown instead.
HLP_SYS_009
What JARVIS can do
No feature matches that filter.
Start here
JARVIS is Playology's voice-driven assistant: a heads-up display that knows your notes, your calendar, the weather and the wider web, takes notes when you tell it to, plays music, rings your phone, and can look at your screen or through your camera. Ask it anything, by voice or by typing.
Try
Talking to JARVIS
Speak or type. Click the glowing core (or the microphone) and talk; press / to type instead. Typed and spoken questions take exactly the same path.
The core shows what JARVIS is doing: calm cyan while standing by, brighter while listening, amber with fast rings while thinking, mint while speaking.
A real British voice. Answers are spoken by a free, self-hosted voice (Kokoro) that sounds the same on every device; if it is unavailable, the browser's own voice takes over. The voice, speed and pitch are all adjustable.
Brief out loud, complete on screen. JARVIS speaks the gist (a sentence or two) and shows the whole answer in the box above the command line, with its sources.
Interrupt any time. Say “stop”, “wait” or “enough” and it stops talking at once.
Never cuts off the first word. Each answer starts with a short lead-in, and the sound output is kept awake while JARVIS thinks, for slow Bluetooth headsets and laptop sound chips.
Knows the date and time with every question, in your calendar's time zone.
Try
The HUD
Six live widgets sit around the core. Click any of them for a detail panel. Every number is real; when a source is not set up or cannot be reached, the widget says so in plain words instead of guessing.
Local time: clock, day and year progress, ISO week, and world clocks (London, New York, Los Angeles, Tokyo, Istanbul) in the panel.
Today: your Google Calendar as a 24-hour ring, the next event with a countdown, today's list and the week ahead.
Weather: current conditions, the next 12 hours, 7 days, sunrise and sunset, wind compass and the moon phase.
JARVIS core: the health of every service (brain, voice, meaning search, calendar, web, sight), server uptime, how long each answer took, and a one-click brain picker.
Claude API spend: this month against your budget (amber from 80%), daily spend for 30 days, spend by model, cost per answer, share of input served from cache, and rate-limit headroom. Only token counts and costs are logged, never your words.
Headlines: news for Playology's topics (Games, Parks & Live, IP & Licensing, Agentic AI), refreshed every 15 minutes; topics and feeds are configurable.
Calls: the last few phone calls JARVIS made, with a one-line summary of each.
The ticker along the top runs one live line per widget: the brain, notes indexed, next event, weather, spend, moon, top headline and voice.
Display settings (the gear): Calm mode with almost no motion (it also follows your system's reduce-motion setting), replay the intro, or open the galaxy. The brain chip, top right, always shows which model is answering.
Works on any screen, from a TV down to a phone, where the page scrolls under a pinned header with the command line fixed at the bottom.
Notes & memory
Your notes are stars. Every note orbits the core, coloured by its group. Hover a star for its name, click it to read it and see its connected notes. The notes an answer came from light up.
The 3D notes galaxy: press Notes galaxy to fly through every note and link in 3D, with the Playology mark at its centre. Drag to orbit, scroll to zoom, click a star to open it, click the mark to come back.
“Remember that …” saves a note on the spot. It is searchable by the very next question, and a new star is born live in the galaxy, with no reload.
Search by meaning, not just words. “Who are our customers?” finds a note that only says “clients”. A small model runs on the computer itself in milliseconds; it is free, needs no key, and your notes never leave the machine. It is combined with plain word matching, so a note you name by title still comes first.
Honest when it doesn't know. If nothing in the notes is close, JARVIS says “the notes are silent on that” rather than inventing an answer.
Add notes simply: drop any Markdown file into the notes folder; the folder becomes its group in the galaxy.
Try
Answers from the web
Internet first. General questions go to the internet; your notes answer questions about your own things (“my”, “our”, “do I have…”) or when a note plainly matches.
Quick, free lookups first: math, units and figures go to WolframAlpha; everything else to DuckDuckGo and Wikipedia, side by side, six seconds at most each.
Then a full web search with Claude's own search and page-reading tools, up to three searches and three pages per question, when the quick lookups have nothing.
Sources on screen. The pages an answer relied on appear as links under it, taken from the search results themselves, so a link can never be made up. JARVIS speaks the answer, never a URL.
Summarised or word for word. Add “verbatim”, “word for word” or “in full” to hear the source's own text; “read that to me verbatim” reads the last answer's source.
Ask for the web directly: “search the web for …”, “google …”, “look it up”. “Near me” searches use your configured location.
Try
Weather
A live forecast for your location from Open-Meteo: free, no key, and only the city and its coordinates ever leave the machine.
Ask in plain words, up to 16 days ahead and for other cities: “Will it rain tomorrow?”, “What's it like in Chicago on Saturday?”.
Imperial or metric, automatically by country or set by you. The moon phase is calculated on the page.
Try
Google Calendar
Read your calendar: today, tomorrow, next week, free time, the next meeting. Events an answer is about appear as cards with a link to open them in Google Calendar.
Add, move or delete events by voice, including recurring series (“every Monday and Wednesday until Christmas”), and change one occurrence or all of them.
Nothing changes until you say yes. Every change is read back first and shown as an amber card with Yes / No. “No”, “cancel”, or anything else you ask drops it; a waiting change never fires later.
The read-back is the change. It is built from the exact request Google will get, never from the model's own description.
Ambiguity is a question, not a guess: if “the meeting on Friday” matches two events, JARVIS asks which.
Time zones handled: “tomorrow at 3” means 3 pm where your calendar lives. Guests are never invited (Google would email them).
Try
The telephone
“Call me” rings your cell with a standing brief: the date, today's calendar, anything waiting for your yes, and the latest activity.
Listens first. It greets you and waits. Say you're busy and it apologises and gives just the headline and two key facts; say hello and you get the full brief.
Answers only from the brief, and repeats any instruction back once, then puts it on the record for the desk. It never claims to have done something from the call.
Relay a message: “call Dave and tell him I'm running 20 minutes late”. JARVIS reads it back first and dials only on your yes, delivers the message in your name, takes a reply, and is truthful that it is an AI if asked. Voicemail gets the whole message.
After every call one line about how it went is spoken on the desk and kept in the Calls panel.
Speakable: amounts, phone numbers, symbols and links are turned into words before the phone says them.
Try
Music & videos
“Play …” finds a song on YouTube Music, preferring the official audio track, and opens it. Ask for a video (“play the video of …”, “I want to watch …”, “on YouTube”) and you get YouTube instead.
“Play that again” reopens the last one. A card with a Play button always appears too.
Not hijacked: “how do I play chess” or “let's play a game” are ordinary questions. No model is asked to pick a song, so it is instant and free.
Try
Screen & camera sight
Share your screen (the screen button) and ask “what am I looking at?”, “is this chart right?”, “proofread this”. One fresh frame is grabbed the moment you ask and shown beside the answer, so you see exactly what JARVIS saw.
Use the camera: “what is this?”, “what am I holding?”, “what kind of plant is this?”, “help me fix this”. The camera opens if needed, waits a moment for you to aim, then looks. Phones start on the back camera, with Flip to switch.
Short memory for problems: while the camera stays on, follow-ups like “OK, what's the next step?” remember the last three exchanges.
Honest about what it can see: if nothing is shared or the camera is off, it says so and sends nothing. One source at a time, a red CAMERA ON badge while live, and it will not identify a person from their face.
Swapping the brain
Change the model by voice (“switch to fable 5.1”, “try on sonnet 5”, “go back to your normal brain”) or by clicking a model in the JARVIS core panel. The chip in the top right shows which model is answering, amber while swapped.
Known models only: Claude Opus, Sonnet, Fable and Haiku, plus an OpenRouter model. A model that isn't on the list is refused with the list of what is, never swapped for the nearest match.
Runtime only: a restart always comes back on the configured model.
Try
Privacy & safety
Your notes stay on your machine; search by meaning runs locally.
Nothing happens behind your back: calendar changes and calls to other people are always read back and wait for your yes.
No made-up sources: links come from the search results, and when JARVIS doesn't know, it says so.
Spend you can see: every API call's cost is logged and shown against your budget, without storing what you asked.
Preflight: one command checks every live chain (brain, voice, notes, web, calendar, weather, phone) end to end and reports pass, fail or warning.
Shortcuts
Click the core
Talk (turns the microphone on or off)
/
Type a question
Esc
Close a panel, the galaxy or this page; skip the intro