The Rabbit R1 arrived in 2024 as a small orange box that promised to end the app. Announced at CES in January and shipped that spring for 199 dollars, designed by the cult hardware studio Teenage Engineering, it sold a clean idea: instead of tapping through apps yourself, you would talk to the R1, and its “large action model” would operate your apps and services for you. The grid of icons was supposed to become obsolete. Within months, the device had proven the opposite of its own pitch.
In May 2024, a developer got the R1’s software running as an ordinary Android app on a Pixel phone, and The Verge confirmed it. The thing sold as a bold new kind of hardware was, underneath, largely an Android app, one that could run on the very phone it claimed to replace. The founder pushed back on the “just an app” framing, but the damage was conceptual, not just technical. The much-hyped large action model turned out to lean heavily on scripted automations rather than some new species of AI. The R1 did little the phone could not already do, and it did that little more slowly.
When the phone already does the job
The reviews were brutal. One widely watched critic called the device “barely reviewable.” Security researchers found hardcoded keys in the code and user conversations logged with no way to delete them. By September 2024, of roughly a hundred thousand people who had bought an R1, only about five thousand were still using it. A redesigned operating system in 2025 read less like innovation than like an attempt to give the surviving owners a reason to keep the thing on their desk.
The R1 is not a companion, and it belongs to a different corner of the market than Maya or Replika. But its collapse teaches the companion category the same lesson from a fresh angle. A separate product, whether a device or an app, survives only if it does something the platform genuinely cannot. “We act across your apps” was never that, because the phone already acts across your apps. The moment the R1’s core promise was revealed to be software the phone could run natively, there was no reason for the R1 to exist as a distinct thing. The category’s deeper problem is usually framed around emotional harm, but it has a plainer cousin: a lot of AI products are betting on a job the platform already owns.
The market’s quiet verdict
This is the same trap that ended the AI Pin and stranded Pi, arriving from the hardware side. When the differentiator is convenience the phone can replicate, or a personality the model can replicate, the product is not standing on defensible ground. How value actually distributes across this market keeps returning the same answer: durability comes from doing something the platform cannot fold in, not from doing a familiar thing in a new shell. It is also why building for a real, lasting need beats building for a demo.
The R1 set out to be the device that replaced your apps. The market delivered a quieter and more cutting verdict. It decided the R1 was an app, and then it decided the app was not worth keeping open.
Sources: Wikipedia, Android Authority, The Verge (2024, Rabbit R1 launch, the Android-app finding, and reviews). TechCrunch (2024, the large action model). Reporting on R1 usage and rabbitOS 2 (2024 to 2025).
