Rabbit turns your computers into a multi-device AI fleet

Most people's AI workflow is a fragmented mess: you brainstorm in a browser tab, manually copy-paste code into your terminal, and constantly shuffle files between your laptop and cloud storage.

Rabbit wants to glue all those pieces together. On September 22, the startup officially released OS3, pitching it as an agentic operating system designed to run your devices through a single chat interface. Instead of juggling apps, you simply tell OS3 what outcome you want, and it handles the rest.

One thread to rule five screens

At first glance, OS3 looks like a straightforward chat window. But underneath, a single Rabbit account can link up to five devices at once, spanning Windows, Mac, Linux rigs, cloud virtual machines, and the Rabbit r1.

Setting up a device requires just a single command to install the local Rabbit agent. Interestingly, Rabbit ditched the usual sidebar of chat sessions. OS3 runs on one continuous conversation thread, remembering past instructions and outputs in the background while shifting work across your screens.

When a task comes in, OS3 determines whether it should execute locally, in the cloud, or across multiple machines. Powered by Rabbit's DLAM technology, the system can autonomously navigate desktop software, read local files, browse the web, and write or debug code.

It also avoids model lock-in. Rabbit added full bring-your-own-key flexibility, allowing users to swap between frontier labs, model routers, or local setups without losing context, skills, or system memory.

Paste a URL, get a new skill

Adding capabilities to AI agents usually involves editing config files, dealing with environment variables, or wrestling with GitHub repositories.

OS3 streamlines this into a simple copy-paste gesture. If you find an agent skill published online, you just paste its URL into the chat window. OS3 automatically provisions and integrates the skill on the fly without manual setup.

On the security side, Rabbit emphasized that the local agent runs directly on your machine without storing or selling your files. While chat logs and memories live on Rabbit's cloud, contextual data sent to third-party model providers is processed per those providers' terms. Sensitive system actions still trigger a local prompt requiring user confirmation.

Users can onboard today at os3.rabbit.tech. Interactions can happen via desktop or mobile web, Telegram, iMessage, SMS, or the r1, with Rabbit's upcoming vibe-coding cyberdeck slated to run OS3 out of the box.

Why it matters

Pivoting from standalone handhelds toward software orchestration is Rabbit's smartest move in months. After the r1's initial launch struggled with flaky cloud automation, leaning into local execution, multi-machine routing, and bring-your-own-key models targets actual user pain points.

For power users and solo builders, OS3 turns tedious multi-device chores into conversational prompts. Imagine pulling data on an always-on cloud VM and dumping the refined spreadsheet onto your home desktop without setting up remote sync tools. It looks well worth testing for workflow automation, provided you are comfortable granting agentic permissions to your OS.

Rabbit's real innovation here is frictionless extensibility. URL-based skill ingestion removes developer friction, and treating the operating system as an agnostic scheduler is the right strategic layer to occupy.

Still, the core challenge remains reliability. GUI-based computer control is notorious for breaking when an unexpected dialog pops up. Coordinating that fragility across five machines at once could multiply failure points. Rabbit has built a compelling vision, but its survival rests entirely on whether OS3 can execute complex tasks without getting stuck halfway through.