Rabbit OS3 launches with task coordination across up to five connected devices
The AMW Read
The public release meaningfully advances Rabbit's previously reported OS3 testing, while reliable multi-device execution could affect how personal agents are built and evaluated.
Rabbit OS3 launches with task coordination across up to five connected devices
Rabbit has released OS3, a browser-accessible personal agent that can run multiple tasks from one continuing conversation. Users can connect their own model service and up to five devices, including computers and personal cloud machines. OS3 can choose where a task runs based on the files and software it needs, or users can specify a device. It also supports interaction through iMessage, Telegram and Rabbit r1. The launch extends Rabbit's closed testing of OS3 reported in August, moving its device-orchestration concept into public use.
The product targets a persistent problem for AI agents: useful work often spans files, applications and accounts that already live on a user's devices. Running on those devices may reduce setup in a separate cloud environment, while coordination across machines could keep some tasks going when another computer is unavailable. The trade-off is operational: work assigned to a sleeping or disconnected device stops until that device is available. OS3's claim is therefore broader than completing a single browser task; it is an attempt to manage context and execution across an existing workspace.
For builders and investors, the test is whether that coordination reliably saves more time than it takes to grant access, check results and correct mistakes. Rabbit says users can retain their memory, skills and device connections when switching supported models, including local models. OS3 itself is free, while users who select cloud models provide their own API keys and pay for usage. Those choices make task completion, handoff reliability and total model cost more useful measures of adoption than a successful launch demonstration.
