Android Meets AI – Deploying AutoGLM to Turn Your Phone into an AI Phone

How to Deploy AutoGLM to Turn Your Phone into an AI Phone

Posted by Byolio on December 14, 2025
🌐 中文 English

ADBkeyBoard GitHub address: https://github.com/senzhk/ADBKeyBoard
AutoGLM GitHub address: https://github.com/zai-org/Open-AutoGLM

Introduction

Recently, the Doubao AI Assistant phone became a sensation across the internet. Its excellent features and performance have garnered widespread attention, with many calling it the future of AI phones. Coincidentally, I recently discovered the AutoGLM project, which can turn a phone into a device similar to the Doubao AI Assistant phone. Therefore, I’ll share how I deployed it and my evaluation.

What is AutoGLM

AutoGLM is an open-source automatic dialogue assistant based on the GLM model from Zhipu AI. It can run on mobile phones and provides functionality similar to the Doubao AI Assistant. Its GitHub address is: https://github.com/zai-org/Open-AutoGLM

Deployment Steps

Environment Preparation

  1. Your computer needs Python 3.10 or higher. Python Windows download
  2. Install ADB and add it to your system’s environment variables. ADB download. If you are an Android developer and have the SDK on your computer, ADB is included by default. Simply add the SDK to your environment variables. You can verify the ADB installation by running the following command in cmd: adb --version
  3. Prepare an Android phone, enable Developer Mode (by tapping the build number repeatedly), turn on USB Debugging mode, and connect the phone to your computer via a USB cable. You can check the connection by entering the following command in cmd: adb devices If a device is listed, USB Debugging mode is successfully enabled.
  4. If you haven’t installed the ADB KeyBoard app on your Android phone before running AutoGLM, you will encounter an error: Checking ADB Keyboard... ❌ FAILED Therefore, you need to install the ADB KeyBoard app, which converts ADB commands into text. Download the APK and install it on your phone. After installation, restart your phone (it may be slow, please be patient). Then, go to your phone’s Settings -> Language & Input, and enable the ADB KeyBoard app as an input method. Finally, re-enable Developer Mode and USB Debugging mode on your phone, and reconnect it to your computer via USB.

Installing AutoGLM

  1. Clone the AutoGLM project using git:
     git clone https://github.com/senzhk/ADBKeyBoard.git
    
  2. Open cmd in the AutoGLM root directory and enter the following commands to install the project dependencies:
     pip install -r requirements.txt 
     pip install -e .
    

Configuring the ModelScope API Service

Here, we’ll use the free service provided by Alibaba’s ModelScope as an example. Use the model AutoGLM-Phone-9B. Register an account, go to your personal homepage -> Access Tokens, and generate an access token to obtain free API-Inference services from ModelScope. Copy the access token.

Using AutoGLM

Now you can use AutoGLM. The command is:

python main.py --base-url https://api-inference.modelscope.cn/v1 --model "ZhipuAI/AutoGLM-Phone-9B" --apikey "your-modelscope-api-key" "###"

Replace your-modelscope-api-key with the access token you just copied, and "###" with the instruction you want AutoGLM to execute, for example: “Open Meituan and search for nearby hot pot restaurants.” Then press Enter to run. If everything goes smoothly, you will see output similar to:

==================================================
💭 Thought Process:
--------------------------------------------------
The user wants to open Meituan and search for nearby hot pot restaurants. I need to:
1. First, launch the Meituan app.
2. Then, search for nearby hot pot restaurants within Meituan.

Currently on the system home screen, I can see some app icons. I need to use the Launch function to open the Meituan app.

According to the allowed app list, Meituan is one of the available apps.


==================================================
⏱️  Performance Metrics:
--------------------------------------------------
Time to First Token (TTFT): 10.027s
Thought Completion Delay:   10.955s
Total Inference Time:       11.085s
==================================================
--------------------------------------------------
🎯 Action to Execute:
{
  "_metadata": "do",
  "action": "Launch",
  "app": "美团"
}
==================================================

At this point, the Meituan app will automatically open on your phone and search for nearby hot pot restaurants. This successfully turns your phone into an AI phone assistant.

Summary

The AutoGLM project can transform a phone into a device similar to the Doubao AI Assistant phone. Deployment is relatively straightforward; just follow the steps above. However, it’s important to note that the AutoGLM project is still in the development stage, and both its features and performance have room for improvement. It often misunderstands complex tasks or fails to execute them, resulting in various errors. Nevertheless, its potential is immense, offering more convenience and enjoyment for mobile phone users.