Rabbit has launched a cloud-based AI agent called OS3 that runs on any Windows, macOS, or Linux PC without dedicated hardware, letting users connect up to five devices, choose their own AI models, and access the agent via desktop, Telegram, iMessage, or the legacy R1.
Short answer: Rabbit has launched a cloud-based AI agent called OS3 that runs on any Windows, macOS, or Linux PC without dedicated hardware, letting users connect up to five devices, choose their own AI models, and access the agent via desktop, Telegram, iMessage, or the legacy R1.
Rabbit, the outfit that first made waves with its R1 handheld, has rolled out an AI agent that doesn’t need its own box to work. The brains behind it are the OS3 “agentic operating system,” which lives in the cloud but runs locally on Windows, macOS, and Linux machines. You can hook as many as five personal devices to a single Rabbit account and plug in whichever AI models you like. Once everything’s set, OS3 figures out which devices, files, apps, and models are needed to satisfy a request, handling the orchestration behind the scenes so you don’t have to sweat the details.
You’re not tied to a single app, either. Rabbit offers a desktop website, plugs into popular messengers like Telegram and iMessage, and still supports the original R1 for anyone hanging onto it. This multi-channel setup lets developers and everyday folks summon the AI from whatever environment feels most natural-whether you’re hunched over a laptop at a desk, chatting on your phone, or tinkering with that legacy hardware.
The shift follows a comment Rabbit’s founder, Jesse Lyu, made to Wired about halting production of the R1. Rather than ditching hardware altogether, the company is turning its attention to an upcoming “cyberdeck” built for vibe-coding, which will also run OS3 when it drops in the coming months. The cyberdeck is pitched as a niche tool for creators who want a dedicated, portable workspace for AI-assisted development, showing that Rabbit still sees a place for specialized form factors even as it leans harder into a software-first play.
On the privacy front, Rabbit says OS3 won’t store, copy, use, or sell your personal data, though it does keep chat logs and associated memories on its own servers to power context-aware interactions. Any third-party AI models you hook into OS3 remain governed by their own privacy policies, so you’ll need to give those a once-over separately. The emphasis on data minimization might resonate with builders who are wary of handing over sensitive code or proprietary info to opaque services.
For AI developers, the release drops the barrier to experimenting with agentic workflows. No need to buy Rabbit’s hardware means teams can prototype multi-device automations using the laptops or workstations they already have, try out different model combos, and scale to five endpoints without extra licensing headaches. Being able to call the agent from messaging apps also opens the door to chat-driven CI/CD pipelines or lightweight automation bots that respond to Slack-style commands.
If you want to give OS3 a spin, start by creating a Rabbit account on the company’s website, link up to five devices, and pick the AI models you’d like to use. Try a few simple tasks-like pulling a file from a linked laptop to your desktop, or kicking off a script via a Telegram message-to see how the system decides which resources to bring into play. As the cyberdeck nears launch, developers who crave a dedicated hardware box should keep an eye on Rabbit’s announcements for early-access chances.
All in all, Rabbit’s move from a proprietary gadget to a cloud-centric, cross-platform agent mirrors a wider trend in AI: delivering powerful, orchestrated capabilities through software that runs on whatever hardware you already own. For anyone building AI-enhanced tools or looking to weave intelligent assistants into existing workflows, this expansion offers a flexible, low-cost entry point that’s definitely worth a look.
OS3 runs in the cloud but operates locally on Windows, macOS, and Linux machines; you can connect up to five personal devices (laptops, workstations, etc.) to a single Rabbit account and use any AI models you choose.
Rabbit says OS3 does not store, copy, use, or sell your personal data, but it retains chat logs and associated memories on its servers to enable context-aware interactions; any third-party AI models you attach keep their own privacy policies, which you must review separately.
Create a Rabbit account on the company’s website, link up to five devices, select the AI models you want to use, then test simple tasks such as pulling a file from a linked laptop to your desktop or triggering a script via a Telegram message to see OS3’s orchestration in action.
The cyberdeck is a portable, niche hardware box aimed at creators for vibe-coding; it will run OS3 when released in the coming months, offering a dedicated workspace for AI-assisted development while Rabbit continues to support its cloud-based agent on existing PCs.
Source: The Verge
UK AI Security Institute and the EvalEval Coalition have published reproducible benchmark results, releasing Evaluation Cards for five core benchmarks-HealthBench, FrontierMath, Humanity's Last Exam, SWE-Bench Pro and Terminal-Bench 2.0-tested on six frontier LLMs and two cyber-focused evaluations, using the Every Eval Ever schema to ensure transparency.
Anthropic’s Opus 5.5 and OpenAI’s GPT-6 Sol and Luna models lower token prices-Opus 5.5 at $4 input and $20 output per million tokens (20 % cheaper than Opus 5) and Sol/Luna at $2/$10 and $0.10/$0.50 per million tokens, roughly half the cost of their predecessors-offering developers reduced AI expenses.
Microsoft announced on September 22, 2026 that it helped dismantle the subscription-based AI-driven fraud platform EvilTokens, which had compromised roughly 12,000 Microsoft accounts across about 10,000 organizations worldwide after appearing on Telegram in February 2026.
NO COMMENTS YET
Comments are open. Have a thought or a question? Share it below.