Experience AutoGLM in the Lybic Cloud Android Sandbox

Open-AutoGLM is an open-source mobile intelligent assistant framework built on the Zhipu-ChatGLM family of models. It can autonomously complete tasks on Android devices through natural language instructions. With Lybic's cloud Android sandbox, you can experience and run AutoGLM directly in the cloud without preparing a local phone or configuring ADB, making it easy to develop, test, and validate Phone Agent workflows.

Currently, the AutoGLM-Phone-9B model is available for download on Hugging Face and ModelScope. The framework supports mainstream apps such as Gmail, Google Maps, WeChat, Taobao, and Meituan, and works in both English and Chinese.

This article explains how to experience the AutoGLM model within the Lybic cloud Android sandbox.

Zhipu-Phone Agent is a mobile intelligent assistant built on AutoGLM that understands screen content multimodally and executes automated workflows to help users finish tasks. The system uses ADB (Android Debug Bridge) to control the device, leverages a vision-language model to perceive the screen, and combines intelligent planning to generate and execute operational plans. Users simply describe their needs in natural language—such as "open Xiaohongshu and search for food"—and the Phone Agent automatically interprets the intent, understands the current interface, plans the next actions, and completes the entire flow. Sensitive operations require confirmation, and the agent supports manual takeover during login or verification code scenarios. It also provides remote ADB debugging, so you can connect to devices over WiFi or the network for flexible remote control and development.

You can now experience the power of Zhipu-Phone Agent in the Lybic cloud Android sandbox.

Unlike the official repository, you should use Lybic’s forked Zhipu-Phone Agent. This fork integrates the Lybic cloud sandbox so you can run automation tasks without a local Android device.

Features

• No local device needed: Run directly in the cloud sandbox
• No ADB setup: Control via API
• Ready to use: Automatically creates and manages the sandbox environment

Install the Agent

Clone the project and install dependencies:

git clone git@github.com:lybic/Open-AutoGLM.git
cd Open-AutoGLM/
pip install -r requirements.txt

Obtain API Keys

Lybic API key

Visit the Lybic website to obtain your LYBIC_ORG_ID and LYBIC_API_KEY.

Large model API key

Choose one of the following platforms to get an online inference API key for the AutoGLM-Phone-9B model:

1. Zhipu BigModel

• Docs: https://docs.bigmodel.cn/cn/api/introduction
• --base-url: https://open.bigmodel.cn/api/paas/v4
• --model: autoglm-phone
• --apikey: Apply for your API key on the BigModel platform

2. ModelScope (MoDa Community)

• Docs: https://modelscope.cn/models/ZhipuAI/AutoGLM-Phone-9B
• --base-url: https://api-inference.modelscope.cn/v1
• --model: ZhipuAI/AutoGLM-Phone-9B
• --apikey: Apply for your API key on the ModelScope platform

Example usage with third-party services:

# Using Zhipu BigModel
python main.py --lybic --lybic-org-id YOUR_ORG_ID \
  --lybic-api-key YOUR_API_KEY \
  --base-url https://open.bigmodel.cn/api/paas/v4 \
  --model "autoglm-phone" \
  --apikey "your-bigmodel-api-key" \
  "Open Meituan and search for nearby hotpot restaurants"

# Using ModelScope
python main.py  --lybic --lybic-org-id YOUR_ORG_ID \
  --lybic-api-key YOUR_API_KEY \
  --base-url https://api-inference.modelscope.cn/v1 \
  --model "ZhipuAI/AutoGLM-Phone-9B" \
  --apikey "your-modelscope-api-key" \
  "Open Meituan and search for nearby hotpot restaurants"

If you prefer to run the model locally, see Model Local Deployment.

Environment Variables

Command Line Parameters

For more command line options, refer to main.py.