The Second Act: Rabbit Pivots from Hardware Hype to Agentic Software

In the fast-moving world of Silicon Valley, where "agentic AI" has become the latest obsession, Jesse Lyu, the founder of Rabbit, finds himself in a familiar, if slightly ironic, position. Last year, the tech industry was abuzz with the launch of the Rabbit R1—a standalone, orange-red handheld device designed to be the ultimate virtual assistant. It was a bold hardware play that promised to offload tasks like booking Ubers or ordering DoorDash to an autonomous AI agent. However, the reality of the R1, which garnered a scathing 3/10 review from WIRED, proved that the world was perhaps not ready for a dedicated device that often failed to deliver on its core promises.

Yet, as the industry pivots toward software-based AI agents, Lyu is positioning Rabbit not as a failed hardware experiment, but as a pioneer that was simply ahead of the curve. This week, the company announced OS3, a cross-platform "agentic operating system" that functions across desktops, mobile apps like Telegram and iMessage, and, if you still have one, the R1 itself. This move signals a significant shift for a company that once argued the future of AI lived in a dedicated pocket gadget.

A Chronology of the Rabbit Saga

The trajectory of Rabbit since its CES 2024 debut has been nothing short of volatile. Following a high-profile launch that promised a revolution in human-computer interaction, the company faced a barrage of criticism regarding the R1’s reliability. The "agentic" capabilities—the ability for the device to perform actions across third-party applications—were often clunky and prone to failure.

As the months passed, the company navigated significant turbulence. A planned expansion into the Indian market was halted by regulatory hurdles, leading to severe financial strain. Reports emerged last year that some employees had gone on strike, alleging months of unpaid wages. During this period of intense scrutiny, Rabbit remained relatively quiet, focusing on iterative software updates—roughly 50 in total—to stabilize the user experience.

Lyu maintains that the company is now on solid ground. He notes that the R1 project, from a pure balance-sheet perspective, was a success. With over 100,000 units delivered worldwide and a return rate under 5 percent, Lyu claims the hardware business maintained healthy margins of 45 to 50 percent, all without the recurring revenue of a subscription model. "Internally, from an engineering perspective, we didn’t make the wrong bet," Lyu asserts. "I think the challenge was really being extremely limited in resources, team-wise and money-wise, compared to other AI studios that raised billions."

Data and the Hardware Pivot

Rabbit’s current strategy reflects a pragmatic realization: consumers are less interested in carrying a second piece of hardware when their smartphones and laptops are already capable of running sophisticated software. The R1 is no longer in production, and existing inventory is being liquidated.

Despite the transition, Lyu insists the device still has a place in the ecosystem. "Today, if you look at our R1 device, it’s completely different from day one," he says. "We’re more open; the device is yours, you decide what to do with the device." This openness includes allowing users to integrate third-party agents like OpenClaw or Hermes AI, effectively turning the R1 into a customizable hub rather than a locked-down proprietary tool.

Lyu still sees demand for the physical R1, occasionally demonstrating fresh orders appearing on his dashboard. However, the company’s future is clearly tied to OS3. For those seeking new hardware, Lyu teases a "Cyberdeck" device scheduled for release in the coming months—a specialized machine designed specifically to harness the power of "vibe coding" and the OS3 platform.

Decoding OS3: The Agentic Operating System

OS3 is not just an update; it is a fundamental redesign of how a user interacts with their digital life. Unlike the R1’s original closed-loop system, OS3 is designed to be a universal agent that sits on top of existing hardware.

How It Works

The system lives in the cloud but utilizes a local "agent node" installed on Windows, macOS, or Linux. This node acts as a bridge between the user’s local files, desktop applications, and the internet. Powered by Rabbit’s proprietary Direct Large Action Model (DLAM) technology, the system can autonomously execute complex tasks, write code, and debug software.

A single user account can link up to five machines, with the OS intelligently deciding which device is best suited to execute a specific task. The interface is a "brain dump" style stream—a continuous chat interface that replaces the traditional fragmented sidebar of current AI chatbots. Users can paste intents, and the system coordinates the necessary models and tools to fulfill the request.

The "Bring Your Own Key" Philosophy

In a move that distinguishes it from competitors like OpenAI or Anthropic, Rabbit has adopted a "Bring Your Own Key" (BYOK) model for OS3. There is no monthly subscription fee; instead, users supply their own API keys from their preferred providers or integrate their own local open-source models. This provides users with more control over their costs and the specific intelligence models they choose to employ.

Real-World Applications and Privacy Concerns

Lyu has been testing the capabilities of OS3 by automating his own administrative bottlenecks. He points to his weekly R1 sales reporting process. Previously, he would manually extract data from phone-based documents and transfer them to a PC-based master spreadsheet. With the OS3 agent deployed on his PC, he can now issue a voice command via Telegram from his phone, and the agent executes the data transfer autonomously.

In a more extreme demonstration of its "agentic" nature, Lyu recently used the system to stand in for him during a work day. While he was occupied at a video shoot, he tasked the OS3 agent with monitoring his Slack channel. The agent, clearly identified as an automated presence, held a 50-minute conversation with an engineer, accessing relevant files on Lyu’s computer to answer technical questions and resolve issues.

Regarding data privacy, Rabbit emphasizes that the local agent node does not store, copy, or sell sensitive user data. While conversations and memory logs are held on Rabbit’s servers, the actual content of the user’s local files remains on their machine. When a task requires model reasoning, the prompts are passed through Rabbit’s infrastructure to the user’s chosen model provider, which then processes the data according to its own privacy policy.

Implications: The Death of the "AI Gadget"?

The launch of OS3 is a quiet, perhaps unintentional, admission that the industry’s initial attempt at a dedicated "AI hardware" gold rush was flawed. Silicon Valley spent much of 2024 attempting to convince consumers that they needed a new, separate device to interact with AI. Meta, for example, recently faced a high-profile standoff with Amazon, which blocked Meta’s "Muse" AI from crawling its site to perform automated shopping tasks.

Lyu recalls the R1’s early struggles with similar virtual machine-based actions and notes with a smirk, "We’ve been through all of this a year and a half ago."

The implication of OS3 is clear: the "AI agent" does not require a new pocket-sized gadget. The best tool for the job is the one the user already possesses. By moving its software to the desktop and mobile environments where people already spend their time, Rabbit is shifting from a hardware-first company to an ecosystem-agnostic agent provider.

Whether this pivot can secure the long-term viability of the brand remains to be seen. The company is currently lean, with a team of only 15 people and $60 million in total funding. However, by embracing an open, BYOK approach and focusing on cross-platform functionality, Rabbit is attempting to shed the "flop" label and establish itself as a serious player in the next generation of computing. If OS3 proves as effective as its founder claims, the story of Rabbit may be remembered not for its orange plastic device, but for being the catalyst that helped bridge the gap between human intent and machine action.