Rabbit, the startup behind the R1 handheld device that failed to make a lasting impression, is launching a standalone AI agent that does not require any proprietary hardware to function. The new system, called OS3, is described by the company as an agentic operating system that runs in the cloud but operates locally across Windows, Mac, and Linux computers.
How OS3 Works Across Devices
The design philosophy behind OS3 is device-agnostic. A single account supports up to five devices, and users can bring their own preferred AI models rather than being locked into a single provider. OS3 determines automatically which devices, files, applications, and models a given task requires, then orchestrates the work across the available resources.
Access to the agent is flexible. Users can interact with OS3 through a dedicated desktop website, through a paired messaging application such as Telegram or iMessage, or through Rabbit's own R1 device if they still have one. The intent is to place the agent where the user already works rather than forcing a workflow change.
The End of R1 and the Road to a Cyberdeck
Rabbit founder Jesse Lyu confirmed to Wired that the company has stopped manufacturing the R1 device. The decision reflects the reality that the original hardware, despite significant marketing attention at launch, did not establish a durable market position. However, Rabbit is not abandoning AI hardware entirely. The company is developing a new device called a cyberdeck, focused on vibe coding, that will run OS3 when it launches within the coming months.
The shift from a general-purpose handheld to a specialized coding device suggests the company has narrowed its thesis about what AI hardware should be. Rather than trying to replace a smartphone or laptop, the cyberdeck targets a specific class of tasks where an AI agent needs tight integration with a development environment.
Privacy and Data Handling
Rabbit makes a direct privacy claim for OS3: the system will not store, copy, use, or sell user data. That said, the company acknowledges that it retains chats and the memories derived from them on its own servers. Additionally, any third-party AI providers linked through OS3 will handle user data according to their own privacy policies, which may differ from Rabbit's stated practices.
The distinction matters for anyone evaluating the system for sensitive work. The agent's core infrastructure is Rabbit's, but the AI models it draws upon belong to other companies with their own data governance frameworks. Users who want a complete picture of their data exposure need to understand both layers.
What This Signals
The R1 episode was a cautionary tale about AI hardware that prioritizes novelty over utility. OS3 represents a more sober approach: deploy the agent as software first, let it run on the devices people already own, and only then build a dedicated piece of hardware once the software platform has proven its value. Whether the cyberdeck succeeds where the R1 stumbled will depend on whether specialized AI hardware for coding genuinely offers something that a laptop and a cloud agent cannot.