Rabbit Launches OS3, Moving Its AI Agent Beyond the R1

The new system can reach local computers from a browser or messaging app, but its promise of hands-off software work depends on users trusting an agent with meaningful access.

By 3 min read
Rabbit Launches OS3, Moving Its AI Agent Beyond the R1
Rabbit Launches OS3, Moving Its AI Agent Beyond the R1

Listen to this story

The audio brief

About 1:22
0:001:22
Read transcript
Rabbit has launched OS3, a cloud-based AI agent that can take a request in a browser, Telegram, or iMessage and carry it out on a linked Windows, Mac, or Linux computer. That is a meaningful change in direction for the company: the agent is moving beyond the Rabbit R1, and onto the machines people already use. Users install a local node, then connect as many as five computers to one OS3 account. Rabbit says the agent can work with local software, files, and web pages, and can write and debug code. It can also decide which linked machine should handle a task, while keeping the interaction in one continuous chat. Users can paste in the public URL of a third-party skill, and OS3 can install it automatically. The tradeoff is access. Rabbit says files stay on the user’s computer, but prompts may pass through Rabbit’s servers and then to a selected model provider. Conversations and memory logs are stored on Rabbit’s servers. There is no monthly OS3 fee, but users must bring an OpenAI or Anthropic API key, or connect a local model. This builds on Rabbit’s earlier DLAM computer-control tool, but makes the computer foothold broader. Rabbit has stopped manufacturing the R1 and ruled out an R2. The next hardware product is a Cyberdeck for vibe coding, planned within months. The immediate test is whether people will trust OS3 with dependable access to the computers where their work happens.

Story brief

3 key points

Rabbit’s OS3 shifts the company’s strategy from dedicated hardware to a cloud agent that can coordinate work across Windows, Mac and Linux machines, plus Telegram and iMessage. Users install a local node, link up to five computers, and bring their own model API keys or local models. Rabbit is no longer making the R1 and has no R2 planned, making OS3’s reliability, data-routing choices and computer-control...

  1. 01

    OS3’s local node can access software, files and web pages, and Rabbit says it can write and debug code.

  2. 02

    Rabbit says prompts may pass through its servers and selected model providers; files remain on the user’s machine.

  3. 03

    There is no monthly OS3 fee, but users supply OpenAI or Anthropic API keys or connect local models.

Rabbit’s next bid to make AI agents useful does not require another device in your pocket. The company has launched OS3, a cloud-based agentic operating system that runs through a desktop browser, Telegram and iMessage, while also arriving as an update for the Rabbit R1. Its practical wager is bigger than a new chat interface: an agent can receive an instruction on one screen and carry it out on a linked computer.

A retreat from the one-device bet

The R1 was built around the idea that a dedicated handheld could handle tasks for its owner. OS3 instead puts Rabbit’s agent on the devices people already use. The R1 can serve as an input and output device for OS3, but Rabbit says it is no longer manufacturing the hardware and has no plans for an R2.

The shift follows Rabbit’s January release of DLAM, a computer-control agent for Windows and Mac. Rabbit described that earlier tool as a way to operate a computer after a user granted screen-sharing access. OS3 extends the premise: rather than using the R1 to control a computer, the company wants the agent to live across a person’s machines and message channels.

A cloud agent with a foothold on the desktop

OS3 runs in the cloud but connects to Windows, Mac and Linux through a local agent node that users install. Rabbit says that node lets OS3 work with local software, files and web pages, as well as write and debug code. One account can link as many as five machines, with OS3 choosing which one should handle a task.

The interface is one continuous chat stream rather than a collection of separate task threads. Rabbit also says users can paste the public URL of a third-party agent skill into that chat, allowing OS3 to install it automatically. That turns OS3 into a coordination layer for outside tools as well as Rabbit’s own software.

No subscription does not mean no tradeoff

Rabbit says OS3 has no monthly subscription fee. Instead, it uses a bring-your-own-key setup: users supply API keys from providers including OpenAI or Anthropic, or connect local open-source models. That approach gives users a choice of model provider, but also places the credentials needed to run the agent at the center of the setup.

The privacy boundary is more complicated than keeping files on a laptop. Rabbit says its local agent does not store, copy or sell data, but relevant prompts can still travel to Rabbit and onward to the selected model provider when OS3 needs reasoning. Users who want the convenience of remote instructions and local execution will need to judge that division of responsibility for themselves.

The next hardware is still ahead

Rabbit has said its next hardware product will be a Cyberdeck, a device for vibe coding that is planned to arrive within months and run OS3. For now, OS3 is available without that future hardware. The nearer test is whether Rabbit can make its stated local-software and coding capabilities dependable enough that people will hand an agent access to the computers where their work already happens.

Sources

  1. rabbit.techFirst major update of 2026 — DLAM, OpenClaw, and a surprise
  2. wired.comRabbit Is Back, This Time With an AI Agent App

Loading discussion...

YOUR READING SPACE

Notifications

Rabbit Launches OS3, Moving Its AI Agent Beyond the R1 | Superpower Daily