Google ARTEMIS brings AI agents to Android testing

Google's ARTEMIS project makes Android devices available to AI assistants and automated test suites through instructions in everyday language. Hosted in Google's GitHub organisation under the Apache 2.0 licence, the software can operate connected phones and emulators, move between apps and collect diagnostic information. GitHub dates the repository's creation to 13 August 2026; its latest commit at the time of checking is dated 12 September. This is an overview of an available project, rather than a claim of a launch today.

ARTEMIS connects mobile interaction to the tools developers already use. Its Model Context Protocol server lets compatible assistants, including Antigravity, Codex and Claude Code, request device actions and inspect results. The project also provides a web console, a command-line interface and a Python client for integration with test frameworks. An Android device needs a connection with USB debugging enabled, or developers can use an emulator.

The system combines accessibility information with visual targeting to identify interface elements. Its Flash profile uses a reactive observe-and-act loop, while Pro adds planning, checks before actions and verification of results. The documentation describes Flash as typically taking three to five seconds per step, but that is a project estimate, not a guarantee for every device or workflow.

A concrete recent change arrived on 11 September: a managed accessibility helper with a UIAutomator fallback. The helper reads screen layouts and can be installed or removed separately. This matters because reliable access to interface structure is part of making automated interactions repeatable.

The makers report more than 99 percent task completion on AndroidWorld, a benchmark covering over 100 multi-step tasks. That figure is a reported benchmark result, not independent evidence that arbitrary business apps will work equally well. The README also acknowledges code developed by Minitap.

For development teams, the potential benefit is a shorter route from a written test scenario to execution, screenshots and diagnostic logs. Teams still need to evaluate their own apps, devices and failure conditions. iOS support, on-device lightweight vision models and native Android Studio integration remain roadmap items; the documented offering centres on Android today.