Google unveils ARTEMIS, an AI agent that automates Android apps via natural language instructions

Google's Pixel-Test-Engineering (PTE) Fusion team has released ARTEMIS, a natural language-based Android automation agent. The tool leverages LLMs to autonomously operate and test apps on real devices or emulators, and the team stated that it achieved a task completion rate of over 99% on the AndroidWorld benchmark, setting a new SOTA.

Key Features and Architecture

ARTEMIS offers two execution profiles.

• Flash Execution: A fast, reactive loop operating in 3–5 second intervals, managing context through asynchronous summarization and visual compression.
• Pro Exploration: A deep reasoning mode that performs UI tree and pixel verification before actions, with an Operator recovering from blocks. It takes approximately 15–40 seconds and can handle complex scenarios with over 100 steps.

It applies multimodal targeting (element index, coordinates, visual fallback) to accurately locate elements across various UI frameworks such as Canvas, Compose, and Flutter.

Developer Workflow Integration

ARTEMIS embeds a Model Context Protocol (MCP) server, enabling direct integration with major AI coding IDEs such as Antigravity, Claude Code, Windsurf, and Cursor. This allows users to perform tasks like Logcat collection, screenshot analysis, and device control within the IDE using natural language. It also provides a Python SDK for integration into pytest or CI/CD pipelines and is available on GitHub under the Apache 2.0 license.