September 30, 2026 · Announcements
Figure 1. You speak; your words appear; the steps run, and it stops before anything is posted.
Here is the thing that bugged us. Computers got faster every year, and we still spend the day operating them: open this, find that, copy, paste, click, send. We were learning the computer's language, the way we always have. Commands, then menus, then taps.
We think the next step is the one where the computer learns ours. You say what you want, the way you would say it to a friend, and it happens. That is AffectiveOS. And Braino, a companion who lives in your notch, is the first thing we built on it.
We grew up watching Tony Stark talk to Jarvis. This is our way of building it, for the Mac you already own.
So what does it actually give your Mac? Five things it never had.
Senses. It sees the screen you are on and hears you, even for an hour straight. Nothing to copy, paste or explain.
Voice. It talks back, out loud. Cut it off mid-sentence and it stops, like a person would.
Hands. Say the task and it gets done in the apps you already use: the message sent, the file found, the form filled. 129 tools, on the real Mac.
Memory. Everything it hears, says and does lands here and stays, on your Mac. Not what you typed into a chat: what actually happened, all day.
Trust. Anything that spends money, sends a message, posts in public or deletes something waits for your yes.
One brain, two speeds. Think of how you work. Some things you just do without thinking, and some you stop and plan. AffectiveOS is the same. A reflex on your Mac figures out what you mean: a chat, a quick action or a bigger job. The easy ones it just does. The big ones go to the thinking, which plans the whole job once. And the reflex learns from the thinking, so what needed a plan yesterday can be a reflex tomorrow.
We buy the thinking, build the hands, and keep the memory. We are not trying to train the biggest model. We use the best ones of the day, from labs like Anthropic and OpenAI, and swap them as they get better. Everything that acts on your Mac, we built. And what it remembers is yours, on your machine.
An illustration, not a measurement: the steps you take by hand for everyday jobs, against saying one sentence.
Figure 2. What each kind of tool covers today.
Honestly, every tool out there does one piece of this well. A chat assistant is smart, but then you do the work. Dictation types faster, and that is it. Cloud agents can act, but on some computer that is not yours. We wanted all of it together, on the Mac you already own.
A general comparison of categories, not of any one product. Categories change quickly; this reflects them as of September 2026.
Figure 3. One request, broken down. Typical, not a benchmark: the hands take as long as the job does.
Most of what we ask for is small, and small things should feel instant. So a reflex on the Mac handles them itself, in about five hundredths of a second, before any big model is even involved. Bigger jobs get planned once, in about two seconds, and then you watch them happen.
Typical times measured in our private beta on a MacBook, from the moment you finish speaking. They vary with the job, the network and the model in use.
Figure 4. Where things live.
Something that sees your screen all day has to earn that. So what AffectiveOS remembers stays on your Mac. Private windows and password apps are never written down, nothing it sees is sold, and you can read, fix or delete anything it keeps, or pause it whenever you like.
Figure 5. AffectiveOS arrives inside Braino, in the notch.
You do not install AffectiveOS on its own. It comes inside Braino, which is in private beta on Mac and Chrome right now. We are letting people in a few at a time. Get early access, or read why we are building this.
Footnotes
Replying to an email in under five seconds is the time for Braino to draft the reply; you still read it and press send.
The comparison table describes categories of tools in general, not any single product.
The memory chart is illustrative and is not based on a measurement.