- English
- 简体中文
Web, PC, Mobile, and more
- Test and automate with natural language across web, mobile, and desktop
- One unified API and test suite — the same way on every platform
- Reaches what selectors cannot — unlabeled elements, canvas, native apps, and cross-origin frames
Web
Add Midscene to your Playwright or Puppeteer tests, or drive your own Chrome via Bridge Mode.
PC
Test and automate desktop apps on macOS, Windows, and Linux with natural language.
Mobile
Test and automate Android, iOS, and HarmonyOS apps on real devices and simulators.
Any Interface
Automate any interface you can screenshot — beyond DOM and accessibility limits.
Vision-driven Capable, cost-effective
- Observe interfaces, perform actions, and verify results as people do
- Use cost-effective vision models to reduce API costs for frequent E2E testing
- Choose from multiple models and combine planning, localization, and page understanding as needed
Doubao Seed
Strong visual grounding for reliable element location — a solid default for Midscene.
Multiple model options
Supports DeepSeek, Qwen, GPT, Gemini, Kimi, and more.
Model API cost
Doubao Seed 2.1 Turbo: $0.59 total API cost for 60 AppControlBench tasks, with 58 passed.
Multi-model combo
Pair a planning model with a vision model to raise task completion rates.
Ready-to-use UI testing toolkit
- Practical APIs to script tests and control automation flows
- Supports extending your own UI action agents
- Lowers the maintenance cost of your UI tests
Rich APIs
Auto-planning for whole flows, plus atomic APIs like aiTap and aiAssert for precise tests.
Skills
Drop-in Skills let AI coding agents test your UI through Midscene CLIs.
Reports & Playground
Replay every step in a visual report, and try ideas fast in the playground.
Flexible Integration
Write tests in YAML with Midscene Test and extend them with custom TypeScript Nodes.
Benchmarks
- Explore Midscene's results on AndroidWorld, MobileWorld, and AppControlBench.
93.1%
