# ARTEMIS

A tool that turns natural-language instructions into Android automation for AI coding assistants and test suites, driving real devices or emulators and capturing logs and screenshots.

ARTEMIS lets AI assistants and test suites drive real Android phones and emulators from natural-language instructions, executing testing workflows and everyday cross-app tasks. It targets UI elements by index when available, falling back to coordinate and visual locating for custom interfaces.

Model Context Protocol (MCP) integration lets IDEs such as Antigravity, Claude Code and Windsurf drive a connected test device and collect Logcat output and screenshots. It runs a reactive observe-and-act loop with asynchronous history summaries, typically taking a few seconds per step, and checks a target before each action, returning blocked actions to the operator for recovery so long-running exploratory and stability tests can continue. It reports a completion rate on Google Research's AndroidWorld benchmark, a set of over 100 multi-step tasks.

A one-click startup script installs system dependencies such as ADB, scrcpy, FFmpeg and Python's uv, and can install global MCP configuration and agent rules into IDEs including Antigravity, Cursor, Claude Code, Codex, Windsurf, VS Code, Cline/Roo and OpenClaw. It is written in Python under the Apache-2.0 license.

Website: https://github.com/google/artemis