Rabbit has launched OS3, a personal AI agent that runs in the cloud and connects to a user’s computers through an installed agent, covering up to five machines per account.
The company is best known for the pocket-sized R1 device, and OS3 does not require it. The agent installs on Windows, macOS and Linux, plus cloud virtual machines and dedicated AI hardware.
Its distinguishing choice is model neutrality with a catch. Users bring their own key, signing up with a provider such as Anthropic or OpenAI or supplying a local model, and can route through aggregators such as OpenRouter. The agent swaps between models without losing context, memory or skills, so cost can follow task complexity.
Rabbit says OS3 imports skills from other assistants by pasting a URL, and it includes guardrails: it acts only on instruction, requires consent for sensitive actions and asks the operating system for elevated permissions locally.
“We’ve spent two and a half years building toward a system you don’t operate, but instead just tell it the outcome,” founder and chief executive Jesse Lyu said.
The category is filling up, with always-on agents arriving from frontier labs and a wave of consumer AI hardware. Rabbit’s answer is to own the orchestration layer rather than the model.
