Rabbit Returns as the App It Tried to Replace

Rabbit, the company led by Jesse Lyu, announced OS3 on September 22 as an AI agent designed to work across screens rather than remain inside its bright orange R1 device. Rabbit says people can reach the software through a desktop browser, Telegram, or iMessage, while existing R1 owners will receive OS3 as an update. The company is therefore expanding the agent without requiring customers to buy its dedicated hardware.

The practical offer remains less complete than the platform list. Rabbit originally promoted agents that could perform errands such as ordering food or booking a ride, but the supplied announcement does not establish which OS3 tasks are ready for routine use, what consumer price applies, or which accounts and personal records the agent can access. Rabbit has previously avoided charging a subscription for its AI features. Buyers still need current terms, not inherited assumptions.

That missing detail matters because the R1 was sold in 2024 as an exit from ordinary app navigation. A person would speak to one purpose-built object, and the object would operate services on the person’s behalf. Reviews found that the central agent functions often failed. The purchase delivered the shape of the future before the service could reliably complete the work.

Lyu now argues that the hardware was financially successful. He says Rabbit shipped more than 100,000 R1 units worldwide, recorded returns below 5 percent, and earned margins of roughly 45 to 50 percent on the device. Those are company claims, not a substitute for performance. Rabbit has raised about $60 million, employs 15 people, and endured financial strain after a planned India launch encountered regulatory barriers.

OS3 makes the original purchase easier to account for. The phone was not an obsolete container after all; it was an installed base Rabbit could not afford to recreate. The orange box is now a receipt with a screen. It can still be useful if the update improves it, but its exclusive purpose has disappeared when the same agent can enter through devices customers already maintain, charge, insure, and replace.

The pivot could nevertheless benefit users. Software distributed across familiar screens can reach more people, receive updates faster, and spare buyers another battery and another object. The useful test is not whether Rabbit anticipated the current enthusiasm for agents. It is whether OS3 can complete a purchase, reservation, or message correctly, disclose when it needs help, and recover cleanly when a third-party service blocks automation.

Prospective users should wait for a task-by-task account of reliability, permissions, pricing, and human supervision. R1 owners need a support timetable, not merely compatibility on launch day. If OS3 works, Rabbit may finally have a service worth using; if the company later sidelines the hardware, its earliest customers will have paid extra to discover that the shortest route to the future was an app.