The day

The chat opens straight away, with the box at the bottom ready for you to type. On a new conversation it shows only the Nirvaana logo and “Build your world.” Type “Catch me up” and Nirvaana answers with what is due and what changed. It is sent as your own message, like anything else you type.

The first time you open it each day on the web or in the Mac and Windows app, it quietly looks over the last day you worked (the daily tidy, below). Nothing is pressed and nothing pops up while it works. When it is done, these can sit at the top of the chat:

• “Last night's check”: up to three short lines about the last day you worked (what's coming up today, one loose end, one small thing to make the day easier), where it is on for you. After a gap it names that day instead, like “Friday's check”. “Thanks” puts it away.
• Your standup, where it is on for you (below).
• A moon forming in one of your worlds, as a suggestion: “Make it a moon” or “Not now”.

Where it is on for you, “Morning note” sends one notification to your phone or browser in the morning, “See what's due today”, only on a day you haven't opened Nirvaana yet. Turn it off in “Settings”, under “Notifications”.

On your very first conversation, a line over the box says “Nirvaana is an AI, not a person. It can be wrong, so check anything that matters.” Every conversation after that has a smaller line under the box: “You're talking to an AI.”

Where you see it, the first time you open Nirvaana each day on the web or in the Mac and Windows app, “Your standup” holds one line for your team, drafted from your own notes since yesterday. Only you see it. Change anything, choose the world under “Post to”, and press “Post”; “Not today” puts it away. Nothing is posted any other way.

Posted lines show on that world's page under “Today's standup”, with “Take back” beside your own. In “Settings”, under “Notifications”, then “Daily standup”, you can turn it off.

Type in the box at the bottom and press Send, or just press Enter. To write a second line without sending, hold Shift and press Enter. The reply is set as plain text, not speech bubbles, because most of it is worth reading properly.

While it works, the Nirvaana mark sits where its reply will appear: the sun rises over the horizon and the rings build. When it uses something, it tells you in your words rather than technical ones: “Kept that.”, “Noted the promise.”, “Looked through what I know.”, “Left a question open for tomorrow.”

For a question that deserves real thought, write it and press “Think harder” instead of Send. That one message goes to Nirvaana's deepest thinking, which takes longer and weighs the other ways of seeing it before it answers. The answer comes back under the reply, in a box of its own.

One message at a time. If you send another while it is still answering, it says “I'm still on your last message.”

The box at the bottom starts with “›”, and what you type there is set in an even, typewriter-like letter. You never need anything special: type the way you would to a person, and press Enter.

• ⏎ send.
• ⇧⏎ new line.
• ↑ ↓ what you asked before, in this conversation.
• / this menu.
• @ one of your sessions on Live.
• esc closes a menu.
• ⌘⏎ (Ctrl+Enter on Windows) sends with “Think harder”.

With the box empty, the up arrow brings back the last thing you asked in this conversation; press it again for the one before, and the down arrow to come back. It never replaces words you are typing. A quiet line of keys under the box reminds you of these.

Type “/” at the start of the box and a menu opens above it, with one sentence for each command. Keep typing to narrow it, use the arrows to choose, Enter or Tab to pick, and Esc to close it. A command only counts when you type it yourself at the start of the box.

• “/world”: lists your worlds; pick one and this conversation is in that world alone, as if you had chosen it in the row of worlds. “/world” and a name switches straight there when only one world fits; when more than one does, it lists them to choose from.
• “/template”: lists the starter templates (a product launch, a team update or a site); pick one and the box starts with its first sentence, for you to finish in your own words and send. Nirvaana plans nothing until you send it, and nothing starts until you press “Approve” on the project card. See Starter templates.
• “/approve”: brings the first card on the page that is waiting for your approval into view. It never approves anything: you still press “Approve” on the card yourself. If nothing on the page is waiting, it says so.
• “/help”: one page of the commands you have and the keys, over the box. Esc puts it away.
• “/history” and words: finds what you asked, what each agent was told, and each agent's summary, in this conversation's world (below).
• “/catchup”: what happened in this conversation's world since your last message there (or in the last day): the agents that finished, with what each wrote back, the ones still working or waiting, and how many cards wait for your approval. Read from what is kept, so it costs nothing.
• “/agents”: lists the agents started from this world that are still waiting or working.
• “/stop”: lists those agents with numbers; “/stop 2” stops the second in the last list shown in this conversation, and “/stop all” stops every one, as each card's “Cancel” would. If that list has changed since, nothing stops and the reply shows it again. Nothing new starts for them, and anything already done is kept.
• “/run” and words: sends one piece of build work to your computer, held to the same test as any message (below). Anything else, it handles here, where what would leave your computer waits for your Approve.
• “/computer” and words: sends them to your computer, whatever they ask (below).

new

Type “@” and the menu lists your own coding sessions running on Live (Claude Code, Codex or Gemini in your own terminal), each with what it is doing. Pick one, then type what to tell it. It reads your words on its next turn, and a card under the reply says when it has. Only a name you pick from the menu counts: typing “@docs” without picking it is just words.

“@api-refactor add tests for the new orders route before you push”

• With two or more running, “@all” sends to every one running when you picked it. A line first names who it goes to; press Enter to send it, or Esc to keep it. One message goes to at most 6 sessions: with more running, the first 6 get it, and that line says how many don't.
• Only your own sessions are listed, never a teammate's, and each still asks its own questions.
• Only words you type go to a session. Anything that looks pasted, and files, stay here, where they are read as data.
• “/stop @api-refactor” takes back what you sent it that it hasn't read yet. The session itself keeps running on your computer: stop it in its own window, or on Live.
• On its card, “Raw log” shows the last steps Live keeps for the session, with a way to open it on Live. Its full log stays on your computer.

new

At the top of the conversation, a quiet line names the world it is in, or “Your world”. While any of your own sessions are working, it also says how many, like “2 agents running”; press it to open Live.

new

Under each reply, a small line says who answered it:

• “Ran on Nirvaana”: Nirvaana answered, from what it keeps for you.
• “Ran on your ChatGPT plan (Codex)”, or your Claude or Gemini plan: the work went to that app on your own computer, and ran on the plan you already have.
• “queued for your computer · nothing has started”: your computer isn't on yet. The work waits, and starts when it is. The card above the line follows it.
• “Ran on Nirvaana · from what's stored”: a list read from what is already kept. Nothing new was written.

new

Ask plainly for one thing to be built or fixed, in one short line (a page, a site, an app, a script, a test, perhaps for or with something), and it goes straight to Claude Code on your computer, in a folder you allowed. You can follow it, and cancel it, in its card. When it is done and its note names a pull request on GitHub, the card has “Open the pull request”.

“fix the typo on the menu page”

Asking in the Nirvaana app on your phone or computer works the same way: build work goes to your computer, and it is one conversation everywhere.

Anything else stays in this conversation, where anything that would leave your computer waits for your Approve: sending, posting, telling someone, deleting, pushing or deploying, anything that uses your connected apps, a message with more than one job in it or that says what the work should then do, and anything you pasted.

If too many tasks are already waiting, or you have sent a lot today, the reply says so in plain words, and nothing is sent.

To send anything else to your computer, say so at the start: “on my Mac:” or “on my Mac,” (or computer, or laptop), “@computer”, “@” and the end of one of your computers' names (“@studio”), or “/computer”. Ending a request with “on my computer” works too. Then your words go to Claude Code on your computer as you typed them, whatever they ask. Add “in” and one of the folders you allowed to choose the folder; otherwise it works where it always does. A question that starts “On my Mac,” and ends with a question mark is still asked of Nirvaana.

“on my Mac, in SantQ, tidy the notes”

Claude Code there works only in a folder you allowed, with its own tools for files, so it can't reach your other apps or accounts. Only words you type go: nothing pasted, no files, one line, at most 2,000 characters, and nothing with a password, a PIN or a key in it. With “/computer” or an “@”, when one can't go, the reply says why and nothing is sent, and with no computer connected yet it says that too. Said in words, it is answered here instead, as before.

new

Type these at the start of a message. Each looks only at the world this conversation is in, or at your own world when it is in none.

“/history pricing”

Finds what you asked here, what each agent was told, and each agent's summary of what it did, newest first. A line from another conversation links to it, and a summary that names a pull request on GitHub links to that too. What was written is quoted as it was written.

“/agents”

Lists the agents started from this world that are still waiting or working, and where.

“/stop all”

Stops the agents “/agents” lists: “/stop” alone lists them with numbers, “/stop 1” stops the first one in that list. The reply names each one it stopped, and says so if one didn't stop.

When Nirvaana thinks something you said belongs in a world you share with other people, it asks you with a small card right under its reply. The newest ones still waiting when you open the day are at the top, under “Waiting for your approval”.

• Approve names who would see it and asks again: “Approve” or “Not yet”.
• “Not this one” keeps it to you, and it won't ask that again.
• The card can also show the words a fact came from, though no suggested fact carries them yet.

Once you decide, the buttons go and a quiet line says “added” or “kept private”.

When it keeps something it says so under its reply: “Kept that.”

On a computer, the side of the day shows what is coming due and who is waiting on whom, across every world you are in. On a phone it is one line at the top of the day, with the counts; tap it to open it.

• “Reminders”: each world's deadlines, questions nobody has answered in a week, and your own promises due this week. The late ones come first, then the soonest.
• “Blockers”: promises between you and someone in a world. “Waiting on you” is what you owe; “You're waiting on” is what you are owed. Each says for how long.
• A world you have brought into the conversation comes first. Press a line to open that world at “Due and waiting”.

Above them, “Now” holds the few that matter most today, each with its reason. Under each: “Done”, “Let go” on a promise of your own (you are not doing it, and that is fine to say), “Not this” and “Snooze”.

Reminders that are no longer true leave the list. Nirvaana takes one off by itself only when you noted the same promise twice, in the very same words and for the very same time, or when it is just the merge of one pull request it names with its repo, merged since in a GitHub repo your worlds follow. When it only thinks one is done, from what you have told it since (on a paid plan), it asks quietly instead: “Probably done?”, with why, and “Close it” or “Keep it”. Nothing closes without your tap. When you tell your AI one is done, it asks you with a card to “Approve”; it never closes one itself. Nobody ever tidies one you owe someone else: only you end those. Under Reminders, “Tidied 3” opens what went, newest first and however long ago, each with why, who tidied it and “Undo”, which puts it back. Nothing is deleted, and nobody is told.

“Checklist” opens your own list: a few things for today, in your words. Press the ring beside an item to mark it done (it closes, and the sun comes up in it); press it again if you pressed by mistake. On a desk, drag a line to move it; on a phone, the arrows beside it. Under the list is a small chat that tends the list and nothing else: “add call Odile”, “move the permit to the top”, “the tyres are done”. It cannot reach your memory, your promises, the web or your computer, and it never opens the main conversation; ask it for any of that and it says so once. The line under its reply (“Added to the list.”) is written from what actually changed.

Tell Nirvaana to do something later, or every so often, the way you would tell a person: “every Friday at 9, send Kim this week's invoices”, or “tomorrow at 3, remind me to call the vet”. It says it back in one line, like “Every Friday at 9 am, I'll ask before sending Kim this week's invoices.” It can be once, every day, every weekday, or one day each week, at a time in your own timezone.

When the time comes, it does not just go ahead. It asks. A card at the top of your day says:

• “It's time: send Kim this week's invoices. Run it now?”
• “Yes” runs it now, in a conversation of its own, as if you had just asked.
• “Not now” brings the card back in an hour. It works once.
• “Skip this time” does nothing until the next time it comes round.

Your yes starts the task and nothing more. Anything in it that would send, post, share or pay still stops and waits for its own yes. If you leave a card unanswered, it goes when the next one comes, so they never pile up. If Nirvaana is on your phone, the lock screen and your watch say only “A task you scheduled is ready for your yes”, never what the task is.

To stop one, say so (“stop reminding me about the invoices”), or find it under “Scheduled asks” at the foot of Automations, where each one shows its next ask with Pause and Delete.

There is no button to end the day. The first time you open Nirvaana on a new day, on the web or in the Mac and Windows app, it does two things in the background with the last day you worked: the day before, or after a weekend or a few days away, the latest day with something in it, up to a week back.

1. It tidies that day's memory. Where it wrote the same thing twice, or wrote something that has since changed, it merges them into one. Nothing is deleted; the old version is marked as replaced.
2. Where the nightly check is on for you, it reads back over that day's conversation, your open promises and its own open questions, and leaves “Last night's check” at the top of the chat.

Each day is tidied and checked once, however many times or on however many devices you open Nirvaana, and only a day that had something in it. The iPhone app does not start it yet; pressing Goodnight there still does.

Everything that matters, and nothing you have to manage:

• Facts it kept. It searches them by meaning, not by exact words.
• Promises, until they are done, broken or let go.
• Questions it left open, which last night's check can bring back.
• How you like to be written to.

What does not carry over is scrollback. There is no sidebar of past chats, on purpose: the promise is that you should not have to go hunting through old conversations for something it should simply know. A day can hold more than one conversation. While a new day's chat is empty, under the Nirvaana logo and Build your world., “Earlier” opens a short list of recent ones, and you can open one.

The thin strip across the top is on every page. On the left is today's date, read from your own device, so it is your day and not a server's. Next to it is the row of worlds: your own world in the middle, and every world you are in around it. The links for the places you have sit at the end, then “Settings”, “Help” and, on a computer, “Sign out”.

When you leave the day for a world's page or your list of worlds, the small line over the date changes to “Back to today”. Press the date, or your own world in the middle of the row of worlds, and you are back in the day. No page under the strip is a dead end.

The “Night” or “Dawn” button in “Settings”, under “Look”, switches between light and dark. It is your choice, not your phone's: Nirvaana never switches on you because the sun went down.