first look at rabbit’s Android agent
a research preview of an AI agent that can use and control Android apps and devices
Our vision at rabbit is to develop a cross-platform general agent system that enables smart, autonomous agents to act on your behalf.
Last October, we shared our progress with the emerging web-agent architecture and launched a generalist web agent, LAM playground, that can navigate the constrained environment of a web browser.
Today, we want to share our latest progress with a generalist Android agent that uses apps and controls devices. In this video, we tasked the Android agent to complete a single action that controls the system-level settings and also multi-step actions that require coordination among different apps. These tasks include:
- change app notifications from the system settings
- send an AI-generated poem on WhatsApp
- search the YouTube app to play a video
- add drink ingredients from a cocktail app to a grocery list in Google Keep
- generate a revenue plan in Google Docs from a business idea in Google Keep and share it with contacts
- download a game app from Google Play and play it
A lot of examples we are showing today present only the core action loop an Android agent completes, and there are a lot of optimizations to be made on speed, intelligence and how quickly it recognizes the states. As we make progress on the Android agent in the coming weeks, we expect significant improvements on speed, accuracy, and intelligence.
In the coming weeks we will be sharing more updates about our upcoming cross-platform multi-agent system. Our Android agent is just one exciting piece of that puzzle.
“The Android robot is reproduced or modified from work created and shared by Google and used according to terms described in the Creative Commons 3.0 Attribution License.”