A year after standalone AI hardware gadgets made headlines with underwhelming debuts, Rabbit is stepping back into the spotlight. Rather than leaning purely on its pocket-sized hardware, the company behind the Rabbit R1 has showcased a new research preview: a generalist Android AI agent capable of operating directly within mobile applications.
While hardware competitors like the Humane AI Pin have collapsed following critical pushback, Rabbit continues to iterate. The company’s latest demonstration shows that it is actively developing software agents aimed at fulfilling the automated interaction promises made during its original 2024 launch.
- --
The Android Agent Research Preview
In a research preview video released on February 19, 2025, Rabbit device team engineers Kevin and Casey demonstrated the software controlling an Android tablet. The initiative reflects a broader industry shift toward AI agents that take direct control of operating systems to perform tasks end-to-end.
Rather than requiring users to manually tap through interfaces, Rabbit’s Android agent translates natural language instructions into on-screen navigation. According to Rabbit, the demonstrated interactions represent the "core action loop" of what an Android agent completes.
+-------------------------------------------------------------+
| Rabbit Android Agent Core Loop |
+-------------------------------------------------------------+
| 1. User natural language prompt |
| 2. Agent parses instructions via Large Action Model (LAM) |
| 3. Automated screen navigation across Android OS |
| 4. Execution of task across individual or multiple apps |
+-------------------------------------------------------------+Demonstrated Capabilities
During the demonstrations and blog updates, Rabbit highlighted a diverse set of tasks carried out across native Android applications:
- System Setting Adjustments: Toggling and reconfiguring app notification permissions inside Android system settings.
- Content Search: Navigating into YouTube to perform targeted video queries.
- Cross-App Data Transfers: Collecting ingredient information from one app and logging it directly into a note inside Google Keep.
- Creative Communication: Generating an AI-written poem and sending it automatically to a recipient inside WhatsApp by locating the contact and targeting the message field.
- App Installation and Interaction: Downloading a game via the Google Play Store and attempting to learn how to play it.
- --
The Road From Hardware to Operating System Control
Rabbit initially captured public attention in January 2024 when founder and CEO Jesse Lyu unveiled the Rabbit R1. Billed as an intuitive companion half the size of an iPhone, the bright orange, walkie-talkie-styled gadget drew more than 80,000 preorders.
Lyu argued that modern smartphones had become bogged down by app-centric ecosystems, proposing instead a natural-language interface driven by Rabbit OS and its foundation: the Large Action Model (LAM). Rather than relying solely on text-generation models, LAM was designed to convert user words into digital actions across various services.
However, upon release, the physical R1 failed to deliver on many of its advertised features, struggling to complete basic actions reliably. Over the following months, Rabbit pushed updates—such as rabbit OS 2 + Creations and the LAM Playground (a web-based generalist agent)—to gradually build out the execution capabilities promised at launch.
The new Android agent acts as a direct continuation of that software development, running automated actions on standard mobile interfaces rather than relying solely on the bespoke R1 hardware interface.
| Feature / Attribute | Rabbit R1 Launch (Early 2024) | Android Agent Preview (2025) |
| :--- | :--- | :--- |
| Primary Form Factor | Standalone orange hardware handheld | Software agent running on Android tablet/OS |
| Core Engine | Early Large Action Model (LAM) | Evolved LAM and Android Core Action Loop |
| Interaction Style | Push-to-talk button, scroll wheel | Prompt-driven automated screen execution |
| Task Scope | Limited built-in services | Cross-app actions (WhatsApp, Keep, Settings, Play Store) |
| Execution Speed | Prone to failure / non-functional | Functional, deliberate app-level navigation |
- --
Technical Nuances and Third-Party Risks
While Rabbit’s recent demonstration points toward functional cross-app automation, the software remains an early work in progress. Observers noted that the agent's actions on the tablet proceed methodically and slowly, emphasizing that real-world generalist control across the fragmented Android ecosystem presents steep technical challenges.
Furthermore, automated agency introduces legal and security complexities. In its documentation regarding third-party agents, Rabbit notes that third-party AI agents are highly experimental and not maintained or developed by the company. Users must set them up independently and assume the associated operational risks, as Rabbit disclaims responsibility for third-party agent performance, privacy, or account security.
By moving into the Android environment, Rabbit is testing whether its Large Action Model philosophy can thrive inside the existing mobile platforms it once sought to replace.
