Anticipy is an AI pendant. It hears what you agree to do out loud and hands it to a browser agent on your computer, which does it. Eight grams of titanium, one LED, no screen.
I'm Omar Ebrahim, I'm 15, and I build it at Anticipation Labs in Vancouver. If you want the product, specs and pre-orders, go to anticipy.ai. If you want to know why a teenager is putting his time into a piece of hardware instead of an app, stay here.
It started because ChatGPT was banned in English class
Not for the reason you'd assume. I wasn't trying to get around it. I was sitting there wondering what would actually happen if an agent could do the whole assignment in a browser, start to finish, the way a person does — open the tabs, read the sources, type into the box. So I wired one together to find out.
It worked better than I expected, and then the interesting thing happened: I stopped caring about the essay. The essay was never the point. The point was that this thing could act. It could sit at a computer and finish something without me watching it.
Which leaves exactly one question, and it's the one nobody was working on. What tells it when to act?
Every wearable listens. None of them do anything about it.
There's a whole category of devices now that record your day and give you a transcript, a summary, a searchable archive of things you said. They're genuinely well made and I've learned a lot from watching how they're built.
But they all stop at the same place. They give you back a better record of what you already know you said. The commitment you made at 2pm is still a commitment you have to remember, sit down, and personally execute at 9pm.
Recording is a memory product. I think the actual product is dispatch.
That's the bet, and I'll defend it as a claim: the transcript was never the valuable artifact. The valuable moment is the two seconds where you say "I'll send that tonight" — because that is the exact instant a task comes into existence, and it's also the instant you stop thinking about it. Everything after that is you fighting your own memory. If something is going to intervene, that's when.
Why hardware, and not an app
The honest answer is that the input has to be where the commitment is, and the commitment is not in a chat box.
Nobody opens a laptop to type "I told Sarah I'd send the deck." You say it to Sarah, out loud, standing up, and then you leave. A phone app requires you to already have noticed the thing you're about to forget — which is the entire problem, restated. The only way to catch it is to be in the room.
So: a pendant. Worn, not carried. No screen, because a screen is a thing you look at, and the whole design goal is that you don't look at it. One LED, so you always know whether it's on. Titanium and 8 grams because if it's heavy or ugly you take it off, and a wearable you take off is not a product.
How it works
The architecture is deliberately boring, because the interesting part is the last step and everything before it should be as dull and reliable as possible.
- The pendant captures audio and sends it over Bluetooth to your phone. It is a sensor and a radio. It does not run the intelligence.
- The phone works out what, if anything, you actually committed to. Most of what you say all day is not a task, and the hard problem here is throwing almost everything away.
- What survives becomes a task with an owner, a deadline and enough context to be acted on.
- The task is dispatched to a browser agent running on your computer, which opens the tabs and does the work — as you, with your logged-in sessions, on the real web.
The reason it runs a browser on your own machine rather than calling a stack of APIs is that the useful work lives behind logins. Your email, your school portal, your bank, the supplier form that has no API and never will. An agent that can only use integrations can only do the ten percent of your life that shipped an API.
What's hard, and what isn't solved
I'd rather write this section than the marketing one.
Knowing what's a commitment. This is the real problem, and it's not a transcription problem. "I'll send that tonight" and "someone should really send that tonight" are one word apart and completely different. Over-trigger and the device is a nuisance you turn off in a week. Under-trigger and it's jewellery.
Consent and everyone else in the room. A device that hears conversations hears people who didn't buy it. I don't think this gets solved by a paragraph in a privacy policy. It gets solved in the hardware and the defaults — visible state, a physical way to stop it, and not keeping what doesn't need keeping. This is the constraint I'm least willing to trade away and the one I get the most pushback on.
Acting without asking, without being terrifying. An agent that checks with you before every step is a to-do list with extra latency. An agent that sends an email you didn't see is a disaster. The line between those is the actual product and I don't think anyone has found it yet, including me.
Battery, in a shape this small. Every gram is a decision. Everything above competes for the same milliamp-hours.
Note on the name
Anticipy is the AI pendant made by Anticipation Labs. There's also a Python package on PyPI called anticipy, and a separate open-source project with a similar name. No relation to either.
Where it's at
Pre-orders are live at anticipy.ai. I'm shipping the first units and writing about it as I go.
What I need most right now is people who've done this before — hardware founders who've been through a first production run, engineers who've shipped BLE firmware that didn't drop packets, anyone who has taken a wearable from a working prototype to a thing in a box. If that's you, I'd genuinely like to hear from you: omar@anticipationlabs.com
What I'm working on this month is on the now page. How I got here — seven projects starting at age 8 — is in the story.