CtrlK
The Rabbit R1 is a handheld artificial intelligence hardware device developed by Rabbit Inc., a company founded by Jesse Lyu. Announced at CES in January 2024 and shipped to customers beginning in April 2024, the R1 is a bright orange pocket-sized gadget designed by Swedish electronics firm Teenage Engineering. Priced at $199 with no monthly subscription, the device was marketed as a standalone AI companion capable of performing tasks on behalf of users through voice commands, powered by what Rabbit called a Large Action Model (LAM) [1].
The R1 generated enormous initial excitement, selling over 100,000 units in pre-orders and driving $10 million in revenue before a single device shipped. However, upon release, the product received overwhelmingly negative reviews from nearly every major technology publication, with critics calling it slow, unreliable, and lacking basic functionality that any smartphone could handle. The device also became embroiled in controversies over its underlying technology, the company's past involvement in NFT projects, and security vulnerabilities that exposed user data [2].
Jesse Lyu, the founder and CEO of Rabbit Inc., has a background in AI hardware that predates the R1. Lyu previously founded Raven Tech, a company that developed AI-powered smart speakers. Raven Tech was acquired by Chinese technology giant Baidu in 2017, and Lyu worked at Baidu before eventually leaving to start a new venture [3].
The company that would become Rabbit Inc. was originally incorporated as Cyber Manufacturing Co. Under this name, the company raised $6 million in November 2021 for a project called GAMA, described as a "Next Generation NFT Project." The company rebranded to Rabbit Inc. approximately two months before the R1's announcement at CES 2024, a timeline that later attracted scrutiny [4].
Lyu secured $30 million in funding for Rabbit Inc. and assembled a team to build what he envisioned as the next generation of AI hardware: a device that could understand and execute complex tasks across applications, going beyond the simple question-and-answer capabilities of existing AI assistants.
The R1's distinctive physical design was created in collaboration with Teenage Engineering, the Swedish consumer electronics firm known for its retro-styled synthesizers, speakers, and audio equipment. Lyu and Teenage Engineering founder Jesper Kouthoofd first met over a decade ago, bonding over a shared interest in vintage synthesizers and product design. Their collaboration began during Lyu's time at Raven Tech, when Teenage Engineering helped design products like the Raven H and Raven R smart speakers [3].
For the R1, Teenage Engineering created a compact, square-shaped device with a deliberate retro aesthetic. The design features a large push-to-talk button, a chunky analog scroll wheel for navigating the screen, and a rotating camera module with a physical privacy shutter. The bright orange color was chosen to make the device visually distinctive and instantly recognizable [1].
The R1's hardware is modest by smartphone standards, reflecting its positioning as a dedicated AI companion rather than a general-purpose computing device.
| Component | Specification |
|---|---|
| Processor | MediaTek MT6765 (Helio P35) |
| Display | 2.88-inch touchscreen |
| RAM | 4 GB |
| Storage | 128 GB |
| Camera | 8 MP rotating module with privacy shutter |
| Battery | 1,000 mAh |
| Connectivity | Wi-Fi, Bluetooth, 4G LTE (nano SIM) |
| Dimensions | 78 x 78 x 13 mm |
| Weight | 115 g |
| Operating System | Android-based (AOSP) with custom Rabbit launcher |
| Price | $199 (one-time, no subscription) |
The device runs on a modified version of Android (Android Open Source Project) with a custom launcher that provides the Rabbit interface. This detail would later fuel the "just an Android app" controversy [5].
The MediaTek Helio P35 processor is a budget-tier mobile chip originally released in 2018, which contributed to the R1's sluggish performance. The 1,000 mAh battery is notably small, even by smartwatch standards, and was a frequent point of criticism in reviews. By comparison, most smartphones ship with batteries ranging from 3,000 to 5,000 mAh [2].
The R1's central selling point was its Large Action Model, which Rabbit described as a new type of AI foundation model. Unlike large language models that process and generate text, a LAM was supposed to understand and execute actions within applications. The idea was that users could ask the R1 to perform tasks (such as ordering a ride, booking a restaurant, or purchasing items online), and the LAM would navigate the relevant apps and websites to complete those tasks autonomously [1].
Rabbit's pitch was that the LAM had been trained by observing humans interacting with applications. By watching how people navigate interfaces, click buttons, fill in forms, and complete tasks, the model would learn to replicate those actions. Users would log into their various services through a "Rabbit Hole" web portal, and the R1 would then be able to act on their behalf within those services.
Detailed technical analysis of the LAM reveals a multi-layered system architecture [15][16]:
Upon release, the LAM's capabilities fell far short of the marketing promises. At launch, the R1 could connect to only four services: Spotify, Uber, DoorDash, and Midjourney. Even with these limited integrations, the device frequently failed to complete tasks successfully. Reviewers found that the R1 was slower and less reliable than simply using the corresponding smartphone apps [2].
Investigations by Coffeezilla and other journalists later revealed that much of the R1's functionality relied not on a proprietary Large Action Model but on hardcoded scripts using Playwright, a browser automation tool. The majority of the device's conversational abilities were powered by OpenAI's ChatGPT API rather than Rabbit's own AI technology [4].
In response to criticism, Rabbit later released the LAM Playground, a web-based tool that allows users to interact with a redesigned version of the LAM. According to Rabbit, the second-generation LAM was "redesigned from the ground up to see and interpret the web differently." Unlike the first-generation system, which was limited to a specific set of pre-trained web applications, the LAM Playground uses an agent system that attempts to interact with any website. Rabbit stated that LAM features had served over 40,000 requests for tens of thousands of customers as of late 2025 [16][17].
However, testing by Tom's Guide found that the next-generation LAM still had significant issues, with the reviewer noting that the system attempted to "gaslight" them by providing incorrect information with high confidence [18].
The R1 was unveiled at CES 2024 in Las Vegas on January 9, 2024, and the presentation generated remarkable enthusiasm. Jesse Lyu's keynote demonstrated the device performing various tasks: ordering food through DoorDash, calling an Uber, playing music on Spotify, and generating images with Midjourney. The combination of Teenage Engineering's appealing hardware design, the accessible $199 price point, and the compelling vision of an AI agent that could act on your behalf drove enormous demand [1][3].
Several technology publications awarded the R1 their "Best of CES" or similar accolades based on the demonstration. The first batch of 10,000 units sold out in pre-orders on the first day, and total pre-orders eventually exceeded 100,000 units, generating approximately $10 million in revenue before shipping began.
The CES presentation was widely covered by technology media, with many commentators drawing comparisons to the early days of the smartphone era. Some analysts suggested that devices like the R1 could represent the beginning of a post-smartphone paradigm, where dedicated AI hardware would replace the app-based model of mobile computing. The timing of the R1's announcement coincided with a broader wave of AI hardware announcements, including the Humane AI Pin, creating a narrative that 2024 would be the year AI devices challenged the smartphone.
Rabbit organized pre-orders into numbered batches, with each batch shipping sequentially:
| Batch | Approximate Units | Shipping Timeline |
|---|---|---|
| Batch 1 | 10,000 | Late April 2024 |
| Batch 2 | 10,000 | May 2024 |
| Batch 3 | 10,000 | May-June 2024 |
| Subsequent batches | 70,000+ | June-August 2024 |
Return rates remained below 5% in the initial batches, though this figure may reflect the enthusiasm of early adopters rather than broader consumer satisfaction [21].
When the R1 began shipping in late April 2024, the critical reception was devastating. Nearly every major technology publication gave the device negative to extremely negative reviews.
| Publication | Rating/Verdict | Key Criticism |
|---|---|---|
| MKBHD (Marques Brownlee) | "Barely reviewable" | Feels unfinished, not ready for wide release |
| Tom's Guide | 1.5/5 stars | Advised people to avoid the device entirely |
| The Verge | Negative | Limited functionality, slow responses |
| Digital Trends | "A mess" | Poor user experience across the board |
| Mashable | "Something iffy" | Questioned the fundamental value proposition |
| The Shortcut | Do not buy | Advised waiting, noted severe limitations |
Common complaints across reviews included:
Several reviewers noted that the R1 could not do anything that a smartphone with ChatGPT installed could not do better and faster [2].
MKBHD's review, titled "Barely reviewable," was particularly influential. Brownlee stated that the device "feels unfinished" and that it was "not ready for wide release," noting that basic tasks like asking about the weather sometimes took 30 seconds or more to return a result. Tom's Guide's 1.5-star review was one of the lowest scores the publication had given any technology product, with the reviewer concluding that there was "no reason to buy this device" [2].
The R1 became embroiled in multiple controversies following its release, compounding the damage from poor reviews.
In one of the most damaging revelations, Android Authority journalist Mishaal Rahman demonstrated that the Rabbit launcher APK could be extracted from the R1 and installed on a standard Android phone, such as a Google Pixel 6, where it functioned essentially the same way. Rahman and his team managed to get Rabbit's OS working on the Pixel using a leaked launcher APK, which he translated directly into a functional Android application. This suggested that the R1's software did not require dedicated hardware and could have been distributed as a regular Android application [5][22].
Jesse Lyu disputed this characterization, telling journalists that "Rabbit R1 is not an Android app" and that "rabbit OS and LAM run on the cloud with very bespoke AOSP and lower level firmware modifications." He argued that the hardware was integral to the experience and that running the launcher on a phone did not replicate the full system. However, the demonstration undermined one of the product's core justifications: that users needed to purchase a separate $199 device rather than simply installing an app on their existing phone [22][23].
The controversy was covered extensively by Tom's Guide, TechRadar, Gizmodo, and other major outlets, with many commentators arguing that regardless of the technical nuances Lyu cited, the practical implication was clear: the R1's core functionality could run on existing smartphone hardware.
In May 2024, YouTuber Stephen Findeisen, known as Coffeezilla, published an investigation alleging that Rabbit Inc. was "built on a scam." The video focused on several issues [4]:
Rabbit responded by stating that CEO Lyu rejected the accusations and that GAMA and Rabbit were "separate ventures under the same corporate umbrella," noting that over 80% of Rabbit's employees joined after the rebranding.
In June 2024, a group of researchers called Rabbitude gained access to the R1's codebase and discovered that the software contained several hardcoded API keys for critical third-party services. Rabbitude first gained access on May 16, 2024, and contacted Rabbit about the vulnerability. However, Rabbit did not take action to rotate the keys until after Rabbitude made a public disclosure over a month later [6].
According to Rabbitude, the exposed keys provided the ability to "read every response every R1 has ever given, brick all R1s, alter the responses of all R1s, [and] replace every R1's voice" [19].
A comprehensive penetration test and subsequent investigations revealed the full extent of the security issues [6][19][20]:
| Vulnerability | Severity | Description |
|---|---|---|
| Hardcoded ElevenLabs API key | Critical | Could allow unauthorized access to text-to-speech service and associated user data |
| Hardcoded Azure API key | High | Exposed access to Microsoft Azure services used for AI processing |
| Hardcoded Yelp API key | Medium | Revealed business search data and potentially user query patterns |
| Hardcoded Google Maps API key | Medium | Exposed location-related query data |
| Hardcoded SendGrid API key | High | Discovered later; potential access to email communications |
| Unencrypted chat logs | High | All user chats and device pairing data logged with no deletion option |
Researchers found that Rabbit had known about the ElevenLabs vulnerability since May 2024 but had not taken action to rotate the API keys. The SendGrid key, belonging to Rabbit's email provider, was discovered subsequently, raising additional concerns about the company's security practices [19].
Rabbit published a security investigation response on June 25, 2024, acknowledging the issues and stating it had taken steps to remediate the vulnerabilities. However, Rabbitude published a second disclosure after Rabbit's announcement, claiming they still had access to the exposed SendGrid key. The delayed and incomplete response drew criticism from security researchers who argued that the company's approach to handling sensitive API keys was fundamentally flawed [20].
In July 2024, an additional privacy issue emerged when it was revealed that all user chats and device pairing data were logged on the R1 with no ability for users to delete them. This meant that anyone who obtained a lost or stolen device could potentially extract the previous owner's personal data [6].
Following the accumulation of negative reviews and controversy, legal challenges emerged. Claims centered on allegations of false advertising and deceptive trade practices, with plaintiffs arguing that the R1's actual functionality at launch fell dramatically short of the promises made during the CES 2024 demonstration. The gap between the polished demo and the real-world product experience formed the basis of these complaints [24].
Despite the disastrous launch reception, Rabbit continued developing the R1's software throughout 2024 and 2025, releasing over 30 over-the-air updates.
Through the second half of 2024, Rabbit addressed many of the most critical complaints. Battery life improved, response times decreased, and new features were added regularly. The Large Action Model, while still imperfect, became more capable of handling complex requests. By late 2024, some reviewers noted that the device had improved considerably from its launch state, though it still struggled to justify its existence alongside smartphones.
In September 2025, Rabbit released rabbitOS 2, a ground-up rebuild of the device's software that represented the most significant update since launch. The major changes included [7]:
The rabbitOS 2 update represented a fundamental shift in the R1's interaction model. Where the original software insisted on a voice-first approach that often felt cumbersome, rabbitOS 2 embraced touch interaction while retaining voice as an option. The card-based tile system gave users a visual way to browse information and access features without needing to remember specific voice commands [7].
Android Police published a reassessment in late 2025 titled "Everyone hated the Rabbit R1. They were right, but now they're wrong," acknowledging that the device had improved significantly through persistent software updates [8].
The response to rabbitOS 2 from early adopters was dramatically different from the launch reception. Technology forums, from Reddit to specialized AI communities, noted the transformation. Users who had relegated their R1 devices to desk drawers reported pulling them out again to try the updated software. While the reception was still mixed, with many users remaining skeptical of the device's long-term value, the sentiment shift was notable compared to the near-universal criticism at launch [8].
The Rabbit R1 is frequently compared to the Humane AI Pin, another AI hardware product that launched in the same period with similar ambitions of replacing or supplementing the smartphone.
| Feature | Rabbit R1 | Humane AI Pin |
|---|---|---|
| Price | $199 (one-time) | $699 + $24/month subscription |
| Form factor | Handheld device | Chest-mounted pin |
| Display | 2.88-inch touchscreen | Laser projector (projects onto palm) |
| Input method | Push-to-talk button, scroll wheel, touch | Voice, touch, gesture |
| Core AI concept | Large Action Model (task execution) | AI-first computing (ambient intelligence) |
| Designer | Teenage Engineering | Humane (ex-Apple team) |
| Launch date | April 2024 | April 2024 |
| Subscription required | No | Yes ($24/month for T-Mobile connectivity) |
| Critical reception | Very negative (improved with updates) | Very negative (product largely discontinued) |
Both products launched to extremely poor reviews and became cautionary tales about the challenges of creating dedicated AI hardware that competes with smartphones. The Humane AI Pin was subsequently discontinued for consumer sales, while the R1 continued to receive software updates [25].
The R1's turbulent trajectory has become a frequently cited case study in discussions about AI hardware products. Several lessons have been drawn from its experience.
The software-hardware gap: Shipping hardware with unfinished software is far more damaging than launching a beta software product, because hardware reviewers evaluate the product as-is and first impressions are difficult to reverse. The R1 demonstrated that the "ship now, fix later" approach common in software startups translates poorly to physical products.
The smartphone benchmark: Any dedicated AI device must convincingly answer the question "Why can't my phone do this?" The R1 struggled to provide a compelling answer, especially after the Android app controversy showed that its software could run on existing hardware.
AI capability overpromising: The gap between the LAM's marketing promises and its actual capabilities at launch severely damaged consumer trust. Demonstrations at CES showed a polished, capable system that did not reflect the real-world user experience.
Transparency matters: The company's NFT background and the technical misrepresentation of its AI capabilities (relying on ChatGPT and Playwright rather than a proprietary LAM) created a credibility crisis that software improvements alone could not fully resolve.
Security fundamentals cannot be afterthoughts: The hardcoded API key vulnerabilities and unencrypted user data logs revealed that basic security practices were not prioritized during development. For a device that handles personal account credentials and sensitive user interactions, this was a particularly damaging oversight [6][19].
Iterative improvement can partially redeem a product: Despite the overwhelming initial negativity, Rabbit's persistent software updates and the rabbitOS 2 overhaul demonstrated that a dedicated team can meaningfully improve a product over time. The question remains whether such improvements can overcome the reputational damage from a poor launch [8].
As of early 2026, the Rabbit R1 remains available for purchase at $199. The device has improved substantially since launch through consistent software updates, culminating in the rabbitOS 2 overhaul in September 2025. The card-based interface, gesture controls, and "Creations" vibe-coding feature represent a significant evolution from the original product.
However, the R1 has not escaped the fundamental challenge that defined its reception: it remains difficult to articulate what the device can do that a smartphone cannot do better. The initial wave of over 100,000 pre-orders has not translated into sustained mass-market adoption, and the active user base appears to be a fraction of the initial sales.
Rabbit Inc. has continued to develop its Large Action Model technology and has signaled interest in future hardware products, with reports indicating a follow-up device may arrive in 2026. The LAM Playground web tool represents an ongoing effort to prove the viability of the action model concept beyond the R1 hardware. Whether the company can overcome the reputational damage from the R1's troubled launch and the various controversies remains an open question.
The R1's story serves as both a cautionary tale about the challenges of AI hardware and, through its persistent software improvements, a modest example of how iterative development can partially redeem a product that launched prematurely.