Rabbit's OS3 agent runs on your laptop — and the R1 is done
Rabbit spent two years as the cautionary tale of AI hardware. On Tuesday it started selling the software instead: an agent that runs on the machines you already own.
Rabbit has launched OS3, a standalone "agentic operating system" that lives in the cloud but executes work through a local agent node you install on Windows, Mac, or Linux — and founder Jesse Lyu told Wired the company has stopped manufacturing the R1, the pocket gadget that made it famous. The R1 was not, by Lyu's account, a commercial failure: he says Rabbit shipped more than 100,000 of the orange-red devices, saw return rates under 5 percent, and held a 45 to 50 percent margin on the hardware, all without a subscription. What it was, in the end, was a bet that people wanted another screen to talk to a model through. They did not.
OS3 is Rabbit's answer to that. One account links up to five machines, including cloud VMs, dedicated AI boxes, and the R1 as a software update — and the system decides which device should run a given task, moving work between them and pulling the files, apps, and skills it needs. You reach it from a desktop browser, from Telegram or iMessage on your phone, or from the R1. Underneath, Rabbit's Direct Large Action Model layer handles local software tasks and can write and debug code on the machine it is installed on. Pricing follows a bring-your-own-key model: no monthly fee, you supply API keys from providers such as OpenAI or Anthropic, or point it at local open-weight models. Rabbit says the platform is open at every layer that matters — model, skills, devices.
The most striking part of the pitch is what Lyu does with it. He describes checking a weekly R1 sales spreadsheet on his phone, then messaging OS3 through Telegram to have the agent on his PC open the file and fold the numbers into his master sheet. On a day of video shoots, he says, he had OS3 stand in for him on Slack: the agent identified itself as his agent, held a roughly 50-minute conversation with an engineer, and read files off his computer to answer questions. That is a more useful demonstration of an agent than most of the launch videos shipping this year, and it is also where the claims get softer. Rabbit says the local agent will not store, copy, or sell your data, and that file contents stay on your machine — but conversations and memory logs sit on Rabbit's servers, and any task that needs model reasoning sends its prompts through those servers to your chosen provider. The privacy boundary is a file boundary, not an air gap.
For a company of 15 people that has raised about $60 million, shipping a cross-device agent layer is a sharper use of the money than a second gadget. The hardware thesis lost; the software one is what agents actually needed — a place to run that is already yours. We covered the offline-first version of this idea in August — Perplexity's Portable Computer runs AI agents entirely offline. Rabbit's upcoming cyberdeck, a vibe-coding machine, is meant to ship with OS3 as its default operating system within months, according to Wired.
What to watch: whether OS3's local node is genuinely doing the work, or mostly relaying prompts to a model provider while the file access stays the interesting part.
Would you hand an agent your work Slack for an afternoon, or is that a line you would rather not cross? Tell us in the comments.
