The company that tried to kill the app just became one.

The Summary

The Signal

Rabbit's pivot from hardware-first to platform-agnostic marks a complete reversal of its original thesis. The R1 launched as a standalone device designed to bypass the tyranny of mobile apps. Now OS3 admits what the market already knew: people don't want another gadget. They want software that works across the devices they already carry.

The technical architecture is more interesting than the narrative pivot. OS3 runs in the cloud but operates locally across multiple operating systems, automatically selecting which devices, files, apps, and AI models to use for a given task. This is the first real glimpse of what Rabbit should have built from day one: orchestration software, not hardware.

"OS3 will automatically determine the devices, files, apps, and AI models it needs to complete a task."

The model-agnostic approach matters more than it looks. By letting users choose their preferred AI models rather than locking them into a single provider, Rabbit is building middleware, not a walled garden. Five-device support per account suggests they're targeting the reality of modern work: laptop, desktop, phone, tablet, maybe a home server. The agent needs to follow you, not the other way around.

What remains unclear is whether OS3 can actually coordinate complex tasks across this distributed setup. Automatic device and app selection sounds simple until you're asking it to pull data from your Mac, run a model on your Linux box, and surface results in Telegram. The demo will matter less than the reliability at scale.

The R1 becomes a footnote, or possibly a dedicated interface for users who want tactile interaction. Rabbit's original hardware play underperformed, but the telemetry from those devices likely informed what OS3 needed to be. Every failed gadget teaches you what people actually do versus what they say they want.

The Implication

If OS3 delivers on cross-device orchestration, Rabbit might accidentally build the connective tissue for the agent economy everyone keeps talking about. The question is execution. Can they ship reliability before competitors with deeper pockets build the same thing? Watch for integrations with workplace tools and whether enterprise buyers start testing multi-agent coordination across OS3 infrastructure.

For builders, the lesson is simpler: distribution beats differentiation. Rabbit learned the hard way that a better mousetrap still needs to work where people already are.

Sources

The Verge AI | Wired AI