Rabbit Ditches the R1 and Ships Its AI Agent Solo

Rabbit, the startup behind the much-criticized R1 gadget, has launched OS3. It’s a standalone “agentic operating system” that doesn’t need Rabbit’s hardware. The Verge AI reports that OS3 runs in the cloud but works locally across Windows, Mac, and Linux machines. Wired was first to report the news. The company has also stopped making the R1, which tells you where it’s placing its bets now.

This is a big change in direction. The R1 was sold as a pocket-sized AI assistant, and it mostly let people down. OS3 is Rabbit admitting that the software was always the real product, and that people already own the devices they want to use.

What OS3 Brings to the Table

Here are the main things Rabbit is promising, according to The Verge AI:

  1. Up to five devices per account: You can link as many as five computers to a single OS3 account. The agent can then work across all of them, so a task can pull a file from your work laptop and an app from your home desktop.
  2. Bring your own AI models: OS3 lets you plug in your preferred AI models instead of locking you into one provider. That matters because model quality changes fast, and nobody wants to be stuck on last quarter’s favorite.
  3. It picks the right tools on its own: Rabbit says OS3 “will automatically determine the devices, files, apps, and AI models it needs to complete a task.” In other words, you describe the goal and the agent handles the routing. That’s the core promise of every agentic system right now, and the hard part is delivering it reliably.
  4. Several ways to reach it: You can use OS3 through a dedicated desktop site, through a paired messaging app like Telegram or iMessage, or through the R1 device if you already have one. The messaging option stands out. Texting your computer a task from your phone is a very low-friction way to use an agent.
  5. Hardware is on pause, not dead: Founder Jesse Lyu told Wired that Rabbit has stopped manufacturing the R1. The company is now building a vibe-coding “cyberdeck” that will run OS3, and Wired reports it’s expected within “months.”

How It Stacks Up

The source article doesn’t compare OS3 directly with rivals, but the context is clear. Anthropic, OpenAI, and others have all released agents that can operate computers, browse, and move between apps. Rabbit’s pitch is being model-agnostic and working across several devices, so it acts as a coordinating layer rather than a single-vendor assistant.

That’s a smart place for a small startup to be. Rabbit can’t win a model race against the big labs. It can try to become the neutral layer that sits on top of whichever models you already pay for.

The Privacy Fine Print

This part deserves a careful read. Rabbit says OS3 won’t store, copy, use, or sell your data. But it also says it will keep your chats, and the memories built from them, on its own servers.

There’s a second catch. Any third-party AI providers you connect to OS3 will handle your data under their own privacy policies, not Rabbit’s. So your real privacy picture depends on which models you plug in. If you’re pointing an agent at local files on five machines, you need to know exactly where those files end up.

What’s Still Unclear

The announcement leaves a few open questions:

  • Pricing: The report doesn’t say whether OS3 is free, paid, or subscription-based.
  • Reliability: Picking devices, apps, and models automatically sounds great. Doing it without mistakes is another matter, and Rabbit’s track record with the R1 invites some caution.
  • Cyberdeck details: Beyond “months,” we don’t have specs, a price, or a firm date.

Why It Matters

What stands out here is how fast the AI hardware story is shifting. First-wave AI gadgets struggled to justify themselves next to the smartphone. Rabbit is now betting that an agent that lives across the laptops and desktops you already own is worth more than another device in your pocket.

If OS3 works as promised, it’s a solid case study in turning a hardware flop into a software pivot. If it doesn’t, it’ll be one more sign that agent reliability is still the hardest problem in the field. Keep an eye on the cyberdeck launch in the coming months, and check The Verge AI’s original report for more details.

Scroll to Top