Guide from the official docs

How to get started with rabbit OS3: account, API key and rabbit agent

Set up rabbit OS3 step by step: sign up at os3.rabbit.tech, add your own model API key or a local model, connect computers, and what is free.

By AideDex desk · published

This guide follows the official docs and announcements, linked inline. AideDex did not test these products and does not score them. How we check facts.

Last checked Oct 24, 2026.

To start with rabbit OS3, open os3.rabbit.tech in a desktop browser, sign in or create a rabbit account, and finish onboarding in one conversation (rabbit newsroom). During that conversation OS3 asks for an API key from a model provider you choose, or you can point it at a model running on your own computer (rabbit support). OS3 itself is free; any cost comes from the model provider (rabbit support). Installing the rabbit agent on a Windows, macOS or Linux computer is the optional next step that lets OS3 work on that machine (rabbit support).

For the status, regions and plan summary, see the rabbit OS3 agent page. This guide covers the setup steps and caveats from rabbit’s own docs.

rabbit announced the general release of OS3 on Sep 22, 2026; it runs in the cloud and reaches your devices through the rabbit agent (rabbit newsroom).

Step 1: create or sign in to your rabbit account

rabbit’s launch post says to visit os3.rabbit.tech “via desktop” to log in or create a rabbit account, then finish onboarding in a single conversation (rabbit newsroom).

If you already own an r1 or used rabbit intern, there is no new account. rabbit says to sign in with “the same rabbit account and the same email address you have always used” (rabbit support). The old web portal was replaced by OS3. Your old r1 journal and intern tasks are not shown as web pages any more; you get them through settings → account → data export, which builds a ZIP that stays available for 7 days (rabbit support).

Without a key you can sign in, look around, export data and delete your account. The first time you ask OS3 to do something, it stops and asks you to register a key (rabbit support).

Step 2: add a model API key (BYOK)

OS3 runs on an API key you bring. You choose the provider and model and pay that provider directly (rabbit support). OS3 asks for a key during setup, and you can add or change keys later in settings → api keys.

Supported providers listed by rabbit on Oct 24, 2026 (rabbit support):

ProviderHow models are chosen
Anthropic (Claude), OpenAIOS3 offers a list it already knows
Google Gemini, OpenRouter, Vercel AI Gateway, Ollama Cloud, Z.ai, Z.ai Coding Plan, Moonshot (Kimi), MiniMax, xAI (Grok)Read from your account after the key is saved
OpenCode GoRead from your subscription after the key is saved
Local modelServed from one of your connected computers
Custom providerAny OpenAI-compatible endpoint

rabbit notes the list can change and “what is in OS3 at the time is the list that counts.”

The steps, as rabbit gives them (rabbit support):

  1. Create an API key with your provider. A restricted key must be able to list models and call the models you plan to use.
  2. In OS3, open settings → api keys. If a key is already saved, choose add provider.
  3. Pick the provider and paste the key.
  4. Choose register key.

OS3 checks the key with the provider before saving. A rejected key, “including one on an account with no credit left”, is refused with the reason. After saving, the key is write-only: OS3 only shows a masked version (rabbit support).

Pick a model, or nothing runs

OS3 splits work into two tiers, small and standard. Under models, each tier has a key, a model and an effort setting. For every provider except Anthropic and OpenAI, “you have to choose a model before OS3 can run anything on it”; until then requests are refused with a message (rabbit support). Each tier can use a different saved key, so you can mix providers.

A saved key cannot be edited. To rotate, add the new key, point your tiers at it, then delete the old one (rabbit support). For general advice on handing credentials to agents, see AI agents, passwords and payments.

Using a local model

The Local model option runs a model on a computer already connected with the rabbit agent, for example with Ollama, LM Studio, vLLM or llama.cpp. It needs a connected computer that is online, the local address the model is served at, and the model’s id. An API key is only needed if your model server asks for one, and OS3 tests the connection before saving (rabbit support). So the agent install in Step 3 comes first if you want a local model.

Where to find that address, per each tool’s docs:

  • Ollama: its OpenAI-compatible endpoint on the same machine is http://localhost:11434/v1/; clients must send an API key value, but Ollama ignores it (Ollama Docs).
  • LM Studio: start the server from the Developer tab or with lms server start; it can serve on localhost or on your network, with OpenAI-compatible endpoints (LM Studio Docs).
  • vLLM: vllm serve <model> starts an OpenAI-compatible HTTP server with /v1/chat/completions and a /v1/models list; chat needs a model with a chat template (vLLM Docs).

What is free and what costs money

rabbit says plainly: “OS3 is free. You pay your model provider according to its own pricing” (rabbit support). rabbit lists three ways to avoid model charges (rabbit support):

  • OpenRouter free models. Which models are free changes over time. OpenRouter’s own FAQ says free models are capped at 50 requests a day in total, or 1,000 a day once you have bought at least 10 credits (OpenRouter).
  • Google Gemini API free tier, with Google’s own usage limits. Google applies those limits per project rather than per key, and the daily request quota resets at midnight Pacific time (Google AI for Developers).
  • A local model on a connected computer, which “avoids provider charges entirely” but needs hardware that can run it. The hardware floors other agents publish for local models are in OpenClaw and Hermes Agent hosting cost.

rabbit also says Anthropic, OpenAI, xAI, Moonshot, MiniMax and Z.ai bill for usage, and Z.ai Coding Plan and OpenCode Go need a subscription. Free tiers and limits are set by each provider, not by rabbit (rabbit support). You can run the small tier on a free model and keep a paid model for standard work. For other agents with free options, see free AI agents; for how other agents meter usage, AI agent usage limits; and for plan prices side by side, the cost page.

Step 3: install the rabbit agent on a computer (optional)

The rabbit agent is “a small program you install on a computer so OS3 can work on it”. It runs on Windows, macOS and Linux, on machines you own or cloud machines you control (rabbit support).

  1. In OS3, open settings → rabbit agents.
  2. Choose register rabbit agent.
  3. Pick macOS, Windows or Linux. OS3 guesses from your browser, so check it matches the target machine.
  4. Copy the install command and run it on that computer: in a terminal on macOS or Linux, in PowerShell on Windows.
  5. The computer appears in the list once it has paired.

The command holds a registration token that expires; start registration again if it runs out (rabbit support).

If you use the OS3 desktop app on macOS or Windows, signing in installs the agent and connects that computer automatically. To stop this, turn off Connect this computer to OS3 automatically in the app’s rabbit OS3 menu (rabbit support).

On a Mac, the agent needs Accessibility (called device control and data access on newer macOS) and Screen & System Audio Recording to see and operate the screen. Use grant on the mac permissions line, enable the agent in System Settings, then refresh (rabbit support).

Device limits differ between rabbit pages: the launch post says one account can connect “up to five devices” (rabbit newsroom), while the support article says “up to 20 computers registered at once” (rabbit support). A connected computer must be on and online; a task running there stops if the machine sleeps or drops off the network. Agents that keep going with the laptop shut are compared in AI agents that work when your computer is off.

Where r1 fits

You do not need an r1. OS3 works in a browser and in the desktop app; r1 is “an optional physical way to reach the same account” (rabbit support). You can also reach OS3 by iMessage, RCS or SMS, or a paired Telegram bot. Each channel keeps its own thread but draws on the same memory (rabbit support). For existing owners, rabbit says “your r1 keeps working”: once the account is set up on the web, press the button and speak (rabbit support).

Safety and data statements

What rabbit states about data and control (rabbit newsroom):

  • The rabbit agent runs locally and “does not copy, store, use, or sell your data.” Conversations and memory are stored on rabbit’s servers; files on your computer are not.
  • Relevant content is processed through rabbit’s servers and sent to the model provider you chose, which handles it under its own terms.
  • OS3 “does not run tasks on its own”. “Sensitive actions require your confirmation”, and you can revoke permissions at any time.

On a connected computer, OS3 works mainly through the terminal. It can open a web page in a browser that “starts with no sign-ins”, and can drive graphical apps with DLAM “with your permission” (rabbit support).

Model training: under settings → account, the help improve rabbit’s models switch “is on unless you turn it off” (rabbit support). Clearing a conversation under data reset does not delete what OS3 remembers. Deleting your account does not remove files on your own computers; rabbit says to uninstall the agent and delete the .os folder in your home directory if you want that machine cleaned up (rabbit support).

More detail is on the rabbit OS3 safety page.

FAQ

Do I need an r1 to use OS3?

No. rabbit says OS3 works in a browser and in the desktop app, and r1 is optional (rabbit support).

Can I use OS3 without paying anything?

Yes, per rabbit: OS3 is free, and OpenRouter free models, the Gemini API free tier or a local model avoid provider charges, within each provider’s limits (rabbit support).

Why does OS3 refuse my requests after I added a key?

For providers other than Anthropic and OpenAI, you must choose a model under models first (rabbit support).

Sources

6 pages
  1. rabbit: rabbit unveils OS3 (newsroom) rabbit.tech
  2. rabbit support: what is OS3? rabbit.tech
  3. rabbit support: how to bring your own key (BYOK) in OS3 rabbit.tech
  4. rabbit support: do I need to pay for an api key to use OS3? rabbit.tech
  5. rabbit support: how to connect a computer with rabbit agent rabbit.tech
  6. rabbit support: how to download your r1 and intern data, or delete your account rabbit.tech

Spotted something wrong or out of date? Report an error · Corrections log