Google open-sources ARTEMIS, an AI agent that drives real Android phones at 3s per step
aigclink · x · 2026-09-22
Google has open-sourced ARTEMIS (8.9k GitHub stars), an agent that turns natural-language instructions into reliable Android automation on real devices: it reads the screen, finds buttons, taps and swipes, and completes cross-app workflows end-to-end, capturing logs, screenshots, replays and reports. It scores 99%+ on the AndroidWorld benchmark.
How it works
- Doesn't rely on fixed element IDs; combines the accessibility hierarchy/XML to read the system UI tree and locate identifiable controls first, falling back to OCR + a vision model for Canvas/Compose/Flutter UIs where tree-based location fails
- Last resort is coordinate-based tapping; in long tasks, old screenshots are compressed into summaries to avoid context explosion
Two modes
- Flash: fast loop, 3–5s per step, observe-and-act
- Pro: planned, 15–40s per step, handles 100+ step long-horizon tasks
Native MCP support means it plugs into Claude Code, Codex and Antigravity, letting AI coding assistants drive real phones for mobile testing and debugging.
More from coding & agent
- Ora's Ranker 3.0 grades whether agents can actually use your site; scores drop 40+ points — yenkel · 2026-11-12
- Gemini API Managed Agents now support self-hosted sandboxes, incl. Vercel — steren · 2026-11-12
- Rayline launches ARC, a per-turn model router claiming 90% cost cuts — testingcatalog · 2026-11-12
- Karpathy's llm-wiki: a pattern for building personal knowledge bases with LLM agents — stochasticchasm · 2026-11-12
- Agent self-improvement with frozen weights: failure traces add just 4pp, blind search gets 60% — burny_tech · 2026-11-12
- Microsoft on agentic schedulers: LLM proposes, deterministic code validates — WirelessLife · 2026-11-12