SkillFoxx

Type a name, a task or a service

Open-AutoGLM

An open phone agent model and framework: it reads the screen and taps through Android, HarmonyOS and iOS apps from a text command

CLIHigh risk

zai-org/open-autoglmAuthor: Z.ai

Install

pip install -r requirements.txt
pip install -e .
adb devices

Run inside the cloned repository. Enable developer mode and USB debugging on the phone first.

This is third-party code. Review the repository files before installing.

What it does

The AutoGLM-based Phone Agent reads the phone screen with a vision-language model, plans steps and executes them. Android devices are driven through ADB, HarmonyOS through HDC and iPhone through WebDriverAgent. The user writes a plain-language task, such as finding a restaurant in an app, and the agent opens the app and completes the flow. Sensitive actions require confirmation, and login or captcha steps hand control back to a human.

Who it is for. For researchers and developers experimenting with agents for mobile apps.

Good fit when

  • You need to automate an in-app flow without an API
  • You are researching GUI-driven agents
  • You want to drive an app on a real device with text commands

Not a fit when

  • You need stable production automation
  • The model must not see screens with personal data

Example request

Open Chrome and find tomorrow's weather forecast

Limitations

For research and learning only. The main model targets Chinese apps; a separate multilingual model covers English. Requires Python 3.10+, a device with debugging enabled and ADB Keyboard on Android, plus a self-hosted model or a BigModel or ModelScope key. The main README is Chinese, with an English version.

How to disable. Turn off USB debugging on the phone and delete the cloned repository and installed package.

Security check

  • The agent controls the phone and can act inside apps
  • Screenshots are sent to the model, possibly a third-party service

README in short

The README introduces Phone Agent, links the AutoGLM-Phone-9B and multilingual models and mentions Midscene.js integration. It walks through preparing Android, HarmonyOS and iPhone: debugging, ADB or HDC, ADB Keyboard. Then it covers dependencies, choosing a hosted service or self-hosting via vLLM or SGLang, and main.py examples. Apache-2.0 licensed.

FAQ

Do I have to host the model?

No, the README lists third-party services that already host it, such as BigModel and ModelScope.

Is iPhone supported?

Yes, through WebDriverAgent; setup is in a separate guide.

Editors’ pick

An MCP server and CLI from Sentry for building, running and testing Xcode iOS and macOS projects from an AI agent

MCP serverMedium risk6.4K
Editors’ pick

Paul Hudson's skill for reviewing and writing SwiftUI for iOS 26: deprecated APIs, data flow, navigation, accessibility, performance

SkillLow risk4.7K
Editors’ pick

Expo's official agent skills: building, deploying, updating and debugging Expo and EAS apps

PluginMedium risk2.5K

An MCP server to drive iOS and Android: simulators, emulators and real devices via accessibility trees and screenshots

MCP serverHigh risk6.7K