google / artemis
An AI agent that turns natural-language instructions into reliable Android automation with 99%+ benchmark success and native MCP integration for IDE-based testing workflows.
星标趋势
AI 分析
项目摘要
ARTEMIS is an AI-powered Android automation framework that converts natural-language instructions into reliable device actions. It features a reactive observe-and-act loop with multimodal model support, native MCP integration for IDE diagnostics, and achieves 99%+ task completion on the AndroidWorld benchmark.
为什么值得关注
ARTEMIS achieves state-of-the-art 99%+ success on AndroidWorld, a notoriously difficult 100+ task benchmark, demonstrating real-world agent reliability rather than just lab performance. Its native MCP server design integrates directly into AI coding assistants (Antigravity, Claude Code, Windsurf), bridging the gap between AI coding agents and physical Android testing.
优势
- Industry-leading 99%+ task completion rate on AndroidWorld benchmark
- Native MCP server integration enabling seamless IDE and AI assistant connectivity
- Reactive observe-and-act loop with asynchronous history summaries achieving 3-5s per step
- Multimodal model flexibility supporting Gemini, Claude, GPT-4o, and Qwen-VL
- Comprehensive diagnostic output with Logcat capture and screenshots
局限性
- No formal releases in the last 6 months, indicating early-stage stability concerns
- Tightly coupled to Android ecosystem, limiting cross-platform applicability
- Requires physical or emulated Android devices and ADB setup, adding operational overhead
使用场景
- Automated end-to-end Android app testing from natural language specifications
- CI/CD regression testing with AI-driven exploratory capabilities
- AI coding assistant extension for debugging and testing on real devices
- Benchmarking and evaluating mobile agent reliability across diverse tasks