Rabbit: LLM-First Mobile Phone
An “LLM‑first” handheld called Rabbit R1, designed with Teenage Engineering, is pitched as a $199 pocket assistant that uses a cloud-based “Large Action Model” to operate apps and websites on a user’s behalf via voice commands. Commenters are split between finding the hardware charming and the agent concept intriguing—especially for simplifying complex app workflows or helping less tech‑savvy users—and doubting its practicality in a world where smartphones could offer the same via an app or built‑in assistant. Major concerns center on privacy and security (handing over credentials to a proprietary cloud), the lack of a sustainable business model without subscriptions, reliance on always‑on connectivity, and a suspicion that heavily polished demos overstate what the first version can really do.
Product concept & capabilities
- Pocket “LLM‑first” assistant that uses a cloud “Large Action Model” (LAM) to drive existing apps and websites (rides, food, travel, music, chat, etc.) on the user’s behalf.
- Users log into third‑party services through a web “Rabbit Hole” dashboard; device then issues actions as them.
- “Teach mode” records user actions and explanations to create reusable automations (“rabbits”), with a planned marketplace for sharing/monetizing them.
- Many commenters say this is essentially a thin client to a cloud service that could, in principle, be “just an app”.
Hardware, design & UX
- Industrial design by Teenage Engineering gets a lot of praise; device compared to Playdate in size and aesthetic.
- Criticisms: looks chunky, may not fit comfortably in pockets; rotating camera is suspected to be fragile; scroll wheel is underused or not demonstrated.
- Primary interaction is push‑to‑talk voice; some keyboard fallback exists (shake to show on‑screen keyboard), but was glossed over.
- Several see it as a “companion” or “Light Phone‑style” minimal device rather than a true phone; basic telephony support is unclear.
Cloud dependence, privacy & business model
- Inference and app control appear fully cloud‑based; commenters agree current hardware and price preclude serious on‑device LLMs.
- Strong concern about giving a third party broad access and credentials to services (Uber, Spotify, Discord, etc.), potential ToS violations, and a “man‑in‑the‑middle” architecture.
- Privacy assurances on the site are viewed as vague; some fear data harvesting or later paywalls.
- $199 one‑time price with “no subscription” is widely seen as financially dubious given GPU and VM costs; theories include VC‑subsidized loss, future subscriptions, cuts from a “rabbit store,” or data monetization. Many expect the service could be short‑lived, creating e‑waste.
Usefulness vs existing phones & assistants
- Supporters: like the idea of escaping app clutter, especially for elderly, non‑technical, or anti‑smartphone users; see value as a home/household assistant or toy; view it as a bold step toward agent‑driven computing.
- Skeptics: argue everything shown is already possible or easier with a smartphone UI; voice is often socially awkward, low‑bandwidth, and unsuitable for complex tasks (vacations, food orders); fear unreliable agents spending money or making mistakes.
Trust, marketing & realism
- Many criticize the website and keynote: heavy animation, poor responsiveness, mandatory long video, hidden or skipped screen states in demos; some think portions are staged or misleading.
- Some see it as visionary but “10 years early”; others call it doomed or even vaporware, expecting Apple/Google to subsume this with improved Siri/Assistant once incentives and tech align.