
What happened
Rabbit Inc. unveiled OS3, a cloud personal AI agent installed via a single command on up to five machines, working with any AI model under a bring-your-own-key model.
Why it matters
Because the agent is model-agnostic, users supply their own model subscriptions and can swap models between tasks, so cost can track how complex each task is.
What to watch
Rabbit says OS3 only acts on user instructions and needs consent for sensitive actions, so the test is whether those guardrails hold in daily use. Watch how quickly device types beyond Windows, macOS, Linux, cloud VMs, dedicated AI machines and the R1 arrive.
WHO IT HITSThis lands on individual power users and IT admins who want an always-on assistant spanning a PC and phone, since OS3 installs on up to five machines and runs in the cloud. Teams already paying for model subscriptions may weigh it against building on a single vendor's agent.
Summaries like this, in your inbox every morning.
Rabbit is best known for the pocket-sized R1 device, and OS3 arrives as its second act — this time as cloud software rather than hardware. The R1 is not required to run OS3, though it does work with it, so the company is widening its reach beyond the device that made its name.
The launch lands in a crowded field. The body notes a glut of personal AI agents and assistants, including OpenClaw, Hermes Agent, Instinct and most recently Muse, all offering autonomous capabilities on consumer devices, long-term memory of goals, and use of tools like email, calendars and file reading and writing. Frontier labs are moving in the same direction: Claude Dispatch lets Anthropic's Windows and macOS app run local agents on a PC even when the user is away, connected by mobile and able to reach local files.
Rabbit's answer is flexibility rather than lock-in. The agent is model-agnostic but requires a user-supplied key, so it can swap between models without disturbing context, memory or skills, letting users match spend to task difficulty — and it can pull in skills from competing assistants by pasting a URL. Whether that openness translates into trust may hinge on the guardrails the company describes, since the argument for an always-on agent rests on it acting only when told.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, LINE, or Slack.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
At Snapdragon Summit 2026, Qualcomm CEO Cristiano Amon hosted Google SVP Rick Osterloh to discuss the past yea…

On September 21, SoftBank launched more than $11 billion in dollar and euro bonds ahead of a $10 billion OpenA…

Advanced Micro Devices crossed a $1 trillion market value for the first time on September 21 after shares jump…

Redis launched LangCache, a managed semantic cache that stores full question-response pairs outside the model…

ChatGPT Voice now runs on OpenAI's new GPT-6 Astra, Sol, and Luna models and can access plugins like email, ca…

YouTube is adding a storytelling assistant to YouTube Studio that checks a channel's past results, suggests pa…
