30-second summary
Rabbit announced the R1 at CES January 2024 — a Teenage Engineering-designed $199 orange handheld that would run a "Large Action Model" capable of learning and executing arbitrary app workflows. Pre-orders hit 100K+ units within weeks. Reviews at launch (April 2024) found the device was a low-powered Android tablet wrapper running Playwright-style browser automations against four pre-integrated services (Uber, DoorDash, Spotify, MidJourney); the LAM claim was not substantiated by independent investigation. A series of security disclosures (hardcoded API keys, exposed customer data) in mid-2024 and the Rabbithole backend failures in late 2024 left the device effectively unusable. The company has not formally shut down as of April 2026 but the product is broadly treated as dead.
The Pitch
"A Large Action Model for everything you do." The CES 2024 keynote promised an AI that would observe your screen interactions, learn the pattern, and then execute it autonomously — a genuinely novel capability for which no benchmark or architecture details were ever published. The industrial design (Teenage Engineering, orange plastic, physical scroll wheel, camera) was the primary narrative driver. Price was $199 with no subscription required initially; a "Rabbithole" cloud service was added mid-2024 to recover recurring revenue. By late 2024 Jesse Lyu's public posts shifted from "LAM is working" to "LAM is hard," without explaining the architectural gap the original pitch had glossed.
Five Causes of Death
Market
The consumer market for a $199 AI companion handheld is narrow — it sits between the Ray-Ban multimodal glasses form factor (more useful because head-mounted) and the smartphone (more capable because you already own it). Rabbit's CES-era 100K pre-order volume demonstrated a real novelty-driven buy impulse but not a durable habit. Once the gap between the promised LAM and the shipped Android-wrapper was visible in every review, the category-creating pitch collapsed into "a worse smartphone with extra steps." The market for expensive AI-novelty hardware shrank dramatically in 2024-2025 as on-device LLMs on existing phones caught up.
Product
The R1 device at launch could not reliably execute the four app integrations it shipped with. Independent security researchers at Rabbitude published evidence in June 2024 that core prompts, API keys, and some customer data were accessible from the unlocked Android root — indicating the "AI OS" was a thin abstraction over commodity Android and that Rabbit's back-end was running off hardcoded credentials. The physical hardware (Mediatek chipset, 4GB RAM, battery under four hours) was underspec'd for any LLM-local inference, forcing all intelligence into the Rabbithole cloud, which became the single point of failure.
Team
Jesse Lyu is a serial founder (Raven H assistant) with a strong consumer-product aesthetic but limited AI-systems-engineering pedigree. The team publicly visible at launch was small (<50 people) relative to the ambition of shipping a novel AI architecture. The absence of AI-research hires who would speak publicly about the LAM under their own names was the tell — a product with a novel model typically has identifiable researchers behind it. The company's engineering culture, from external observation, optimized for industrial-design press cycles more than for the research required to validate the pitch.
