ARTEMIS is Google’s open‑source mobile automation framework that lets AI assistants and test suites drive real Android devices through natural‑language instructions. The system provides two execution profiles: **Flash** (a reactive observe‑and‑act loop completing each step in 3‑5 s) and **Pro** (a planning‑and‑verification workflow that can span 15‑40 s per step and supports long‑horizon, multi‑agent graphs). Cross‑app automation uses multimodal targeting—preferring element indices when available and falling back to coordinate or visual locating for custom UIs such as Canvas, Compose, or Flutter. IDE diagnostics are exposed via a native Model Context Protocol (MCP) server, enabling tools like Antigravity, Claude Code, and Windsurf to issue commands, collect Logcat output, and capture screenshots. A lightweight accessibility helper is installed on the first task to read screen layout without UiAutomation; if it fails, ARTEMIS falls back to UIAutomator2 and reports the incident. The framework integrates a Python SDK with zero runtime dependencies, supports structured assertions, and can be invoked from the CLI (`uv run artemis run`) or a web console (`uv run artemis ui`). Benchmark results show **99 %+ task completion** on Google Research’s AndroidWorld suite, covering over 100 multi‑step tasks across 20+ applications. The project also includes Android Studio plugins, on‑device lightweight vision models, and plans for iOS and voice‑driven interaction.