Your AI Operator for Web, Android, Automation & Testing
Joined April 2024
- Tweets242
- Following7
- Followers801
- Likes66
Midscene retweeted
Midscene Action 支持用 ChatGPT 登录跑 CI 了。
把授权凭据存进 GitHub Secret，就能让 GitHub Actions 用 ChatGPT 跑 @midscene_ai 回归。会话续期由 Action 处理，不用每个项目再写一套脚本。
Tapilot 已经跑通：每次合并 PR，CI 自动安装 APK，执行固定的自然语言用例。
用例随着代码改动反复跑，发现过的问题也能补进去。出了问题，可以在 Midscene 报告里回看操作步骤和截图。
想接入的话，可以参考 Tapilot 的完整 workflow，包含 ChatGPT 授权和 Cloudflare Workers 报告发布。
Tapilot 示例：
github.com/tapilot-ai/tapilo…
Midscene Action：
github.com/web-infra-dev/mid…
Already running in Tapilot: every merged PR triggers CI to install the APK and run reusable natural-language tests.
When something breaks, the Midscene report lets us replay the steps and inspect screenshots. Each bug fix can become a test for future runs.
Want to try it? Tapilot's workflow includes ChatGPT auth and report publishing to Cloudflare Workers.
Example:
github.com/tapilot-ai/tapilo…
Midscene Action:
github.com/web-infra-dev/mid…
Midscene retweeted
大家平时都是怎么用 AI 做回归测试的？
最近给 Tapilot 接上了 Midscene Test，把核心流程写成固定的自然语言用例。每次合并 PR，CI 都会自动安装 APK 跑一遍，出了问题可以回看操作步骤和截图。
对 Tapilot 来说，最实用的是这些用例能一直留着，随着代码改动反复检查。发现过的问题也能补进去，后续回归时继续测。
这次 11 个用例全过，零重试，完整 CI 跑了 9 分 08 秒。
基于 Midscene 的 tapilot 也接入了 Midscene Test，舒服舒服。
Midscene retweeted
I open-sourced Tapilot 📱
Give your Android phone a task in plain language. No connected computer required.
Built with Midscene. Optional Skills + Codex integration.
Demo: tapilot.lifeos.md
Code: github.com/tapilot-ai/tapilo…
🤖 Made with AI
Midscene retweeted
Tapilot 开源了 📱
不连接电脑，直接打字或说出任务，让 AI 看屏幕操作安卓手机。
基于 Midscene，无需 ADB 或 root，执行过程可回放。也有 Skill／Codex 插件。
源码：github.com/tapilot-ai/tapilo…
官网：tapilot.lifeos.md
终于开源了，拖延了好久 @dingyi 😄
一句话操控 android 上的任何 App，微信也不在话下！
最近 @zhongerxin 的 iPhone use 非常火，我也手痒痒，做了个语音控制安卓手机的 App，底层用的是 Midscene 纯视觉方案，效果见视频！
不用 ADB，不用开发者模式，也不用插线，随时拿出手机，说说想做什么就行。
@dingyi 这个 App 也支持通过 Codex 插件无线控制，效果见配图！
接下来打算开源，想玩的可以关注我。
后面还会基于 Midscene 做更多有意思的产品，也欢迎给 Midscene 点个 star：
github.com/web-infra-dev/mid…
顺便推荐 iPhone use： github.com/zhongerxin/iPhone…
Meet Midscene Test: an E2E testing framework built for natural language. Write your test cases as plain-language steps that AI runs by looking at the screen, then plug in your own business logic as Nodes to make it your team's own E2E framework. 🧪
midscenejs.com/midscene-test…
Midscene retweeted
用 PaperDIY 做了这么个电子墨水监控屏。其中最耗时的部分是找 IT 审批 IoT 设备入网工单....
#M5StackNew 🎉 PaperDIY Assembly & Complete Kit Released
PaperDIY Assembly & Complete Kit are DIY e-paper development board kits, including main control board, 140mAh lithium battery, e-paper display and acrylic screen pressure plate, suitable for building desktop information displays, electronic labels, maker DIY projects, and embedded hardware development experiments.
✨Features
✅ ESP32-S3R8 core controller (16MB Flash, 8MB PSRAM, 2.4 GHz Wi-Fi)
✅ 4.7" e-paper display, 960×540 resolution, 16-level grayscale display
✅ Onboard DC 3.7V battery connector, Low-power power management design
✅ 140mAh lithium battery, microSD card slot
* PaperDIY Assembly Kit: assembly completed by the user, PaperDIY Complete Kit: includes a pre-installed 3D-printed enclosure, with hardware assembly completed at the factory.
🛒 Get Yours:
Official Store:
PaperDIY Assembly Kit: bit.ly/4jYssSu
PaperDIY Complete Kit: bit.ly/3TW9Lo3
AliExpress:
PaperDIY Assembly Kit: bit.ly/4i79YNj
PaperDIY Complete Kit: bit.ly/4s4mLEw
#M5Stack #M5StackNew #IoT #EINK #EPaper #DIYElectronics #Makers
Usage docs: midscenejs.com/model-common-…
Midscene retweeted
I built a voice-controlled Android app with Midscene—even WeChat works.
No ADB, developer mode or USB cable. Wireless Codex control too.
21s demo + screenshot. Follow for beta + open-source updates.
Star Midscene: github.com/web-infra-dev/mid…
cool(๑•̀ㅂ•́)و✧
一句话操控 android 上的任何 App，微信也不在话下！
最近 @zhongerxin 的 iPhone use 非常火，我也手痒痒，做了个语音控制安卓手机的 App，底层用的是 Midscene 纯视觉方案，效果见视频！
不用 ADB，不用开发者模式，也不用插线，随时拿出手机，说说想做什么就行。
@dingyi 这个 App 也支持通过 Codex 插件无线控制，效果见配图！
接下来打算开源，想玩的可以关注我。
后面还会基于 Midscene 做更多有意思的产品，也欢迎给 Midscene 点个 star：
github.com/web-infra-dev/mid…
顺便推荐 iPhone use： github.com/zhongerxin/iPhone…
Midscene retweeted
Midscene supports these capabilities too. As early as last year, community developers had extended it to control a robotic arm for testing in-car infotainment systems.
teapawt.net/midscene_ai/status/200…
Midscene, running directly on an Android device 📱
This community-built app runs natural-language and YAML automation on the phone, with reports you can inspect there. No PC connection needed.
Now featured in Awesome Midscene. Built by lhuanyu 👏
github.com/lhuanyu/midscene-…
A super fast Jev test run ⚡
Watch Jev × Midscene Test in action with this community-built jevAct runner. It also works with an existing Playwright page.
Now featured in Awesome Midscene. Built by KiritoKing 👏
github.com/KiritoKing/midsce…
Midscene retweeted
我把 Midscene 接进了 Doubao Say 的 CI。
AI E2E 在 Ubuntu 和真实 Omarchy 桌面跑，直接操作 GTK、Hyprland 和系统菜单。
用例都带节点截图、AI 判断和回放报告。
我也在维护 Midscene，就拿 Doubao Say 来 dogfood：能不能测真 Linux 桌面，跑一遍 CI 就知道了。
github.com/web-infra-dev/mid…
Midscene retweeted
我们一直在思考：AI 自动化测试的第一性原理是什么？
答案是——纯视觉。 人类怎么操作界面，AI 就怎么操作。
现在，Midscene 全平台操控 Skills 已上线 clawhub.ai：
iOS · Android · Web · Mac · Linux · Windows！ 快让你的 @openclaw 用这一套方案，控制所有平台吧！
What stands out isn't just the rank — it's the setup.
Most agents ranked above us stack multiple models or add an accessibility tree. Midscene runs on a single Gemini-3.5-Flash, acting from screenshots alone. No DOM, no a11y tree.
It's open-source, and we published the full per-task reports — round by round, pass and fail.
Run config: Gemini-3.5-Flash · Midscene v1.9.5 · DeepThink on
→ midscenejs.com/zh/android-wo…