Google Open-Sources ARTEMIS, an AI Agent for Android E2E Automation

Google's Pixel Test Engineering (PTE) Fusion team has released ARTEMIS, an open-source agent tool that turns natural-language commands into reliable Android control. The repository is available at google/artemis on GitHub.

Image Credit: google/artemis (Apache-2.0)

ARTEMIS runs end-to-end (E2E) automation by executing test scenarios described the way a person would say them as real device interactions. It supports the Model Context Protocol (MCP), so it connects directly to modern AI coding assistants such as Antigravity, Codex, and Claude Code, where teams can run E2E workflows and collect logs for debugging.

What the tool does

ARTEMIS bridges the gap between a natural-language instruction and concrete Android device control. Describe the flow you want to test, and the agent observes screen state and performs taps, text entry, and navigation in sequence.

Its standout capability is cross-app automation spanning multiple applications. A single workflow can, for example, look up directions in Google Maps and then play a video on YouTube — scenarios that cross app boundaries without splitting into separate scripts.

According to Google, ARTEMIS achieves over 99% task success on the AndroidWorld Benchmark. For Android teams still relying on repetitive manual QA or brittle scripted tests, that figure makes it worth a serious look.

Who it is useful for

The primary audience is test engineers and app developers responsible for Android E2E and regression testing. Writing test cases in natural language lowers the cost of authoring and maintaining scenarios.

It also fits well into workflows that already use AI coding assistants. Through the MCP connection, Antigravity, Codex, and Claude Code can invoke ARTEMIS workflows directly and pull back debugging logs on failure, keeping the run-fail-inspect loop inside one environment.

Teams that validate flows across apps — directions, search, media, and payments in a single journey, for instance — may find it more practical than single-app automation tools.

Adoption path and verified limits

The canonical source is the google/artemis repository on GitHub, and running it requires a physical Android device or an emulator environment. adb access and MCP connection permissions are prerequisites; the repository's official documentation is the safe reference for exact setup steps.

Its verified limits deserve equal attention. UI changes in the target apps or unexpected popups may require exception handling in the test environment. Because it executes multi-step agent runs from natural-language instructions, timing issues from device responsiveness and network latency should also be checked. Before gating an important release on it, re-measure the success rate on your own representative devices and network conditions.

Sources

• GitHub: google/artemis
• @github_skydoves (X): ARTEMIS announcement post