Rabbit made headlines two years ago with the R1, a small orange handheld meant to replace your phone apps with voice commands. Reviews were brutal. Now the company has quietly stopped manufacturing that device altogether and is betting everything on software instead.
The new product is called OS3. It is not a gadget. It is a program you install on a Windows, Mac, or Linux computer, and it works through a chat window on a website, through Telegram, or through iMessage. You tell it what you want done, and it figures out which of your devices, files, or apps it needs to touch to finish the job.
Founder Jesse Lyu gave an example: a weekly sales report lands in his email, and normally he has to sit at his computer to copy the numbers into a master spreadsheet. With OS3 running on his PC, he can send a message from his phone and the agent does the copying for him. In another case, he had the agent sit in for him on a work chat for nearly an hour while he was busy, telling the other person up front that it was an AI standing in.
This pivot matters more than it looks. Rabbit is quietly admitting that the whole idea of a separate AI device in your pocket, the thing that made it famous at its 2024 launch, did not work out. It is not alone. Humane, the company behind the similarly hyped AI Pin, shut that product down last year and sold its technology to HP for a fraction of the money it had raised. The lesson from both companies is the same: people already own a phone and a laptop, and they do not want a third device just to talk to a chatbot.
What is left standing is the software idea underneath: an AI agent that can act on your existing computer without needing a programmer. That part is worth watching, because it is not unique to Rabbit. Bigger companies are chasing the same goal. Meta just launched a shopping agent called Muse, and Amazon promptly blocked it from its site, saying an unauthorized program moving through customer accounts was not something it would allow. That fight is a preview of what is coming: platforms deciding, case by case, whether an AI agent acting on your behalf is welcome or treated as a threat.
For a small or mid-sized business, the appeal is obvious. An agent that can open a spreadsheet, cross-check numbers, or reply to routine messages saves time, and Rabbit's approach of letting you plug in your own AI provider account instead of paying a subscription keeps the cost predictable. The catch is just as real. Handing a piece of software the keys to your files and messaging accounts means trusting both the software maker and every AI provider connected to it. Security researchers tracking this category flag stolen access keys and hidden manipulated instructions in documents as the two biggest risks of letting agents act freely on company systems. Before adopting a tool like this, ask exactly what it can touch, and who is watching what it does with that access.
Rabbit itself remains tiny, with around fifteen employees and roughly sixty million dollars raised in total, competing against firms with far deeper pockets building the same kind of agent. Its survival is not guaranteed. But the direction it is chasing, software that runs your existing computer instead of needing new hardware, is very likely where the rest of this market ends up too.