Day 3: Testing AutoGLM
Yesterday, I published an article on AI agents, and many people asked:
“I understand the theory, but how do I actually use it?” “Are there any real test cases?” “Can a beginner really get started?”
Honestly, I had the same questions.
To clarify, I spent two days personally testing three deployment methods of AutoGLM.
I encountered several pitfalls but successfully got my phone to operate itself.
Today, I’m sharing this practical tutorial with you.
No fluff; every step has been personally verified.
Why Choose AutoGLM?
There are many AI agent tools on the market, but I chose AutoGLM for a simple reason:
It genuinely allows your phone to “operate itself.”
Not a simulation, not a demonstration—watch your phone screen as the AI opens apps, clicks buttons, inputs text, and completes tasks.
It’s like having an invisible assistant working on your phone screen.
And it’s completely free.
Open-source on GitHub, it can run locally without spending a dime.
Three Deployment Methods: From Beginner to Advanced
I tested three methods, ranked from easiest to hardest.
I recommend starting with the first method and progressing to the others once you succeed.
Method 1: Cloud Version (5 Minutes)
Target Audience: Everyone, especially those with no technical background.
This was the first method I tried, and it’s the simplest.
My Process:
Step 1: Open the official website
Visit this URL:
https://www.zhipuai.cn/autoglm
Step 2: Register an account
I registered using my phone number; the process was simple:
Enter phone number, receive verification code, set password, log in.
Step 3: Experience the cloud phone
After logging in, you’ll see a virtual phone interface.
It’s literally a “cloud phone” that you can operate via the web.
My first test:
I tried entering the simplest command:
“Help me open WeChat.”
Then, a magical scene occurred—
I watched as the virtual phone screen lit up, the WeChat icon was clicked, and the app opened.
It really worked!
Step 4: Advanced testing
After tasting success, I continued testing:
“Help me search for milk tea on Meituan.”
The cloud phone automatically opened Meituan, entered “milk tea” in the search box, and displayed a list of stores.
I also tried:
“Help me search for AI popular science on Xiaohongshu.”
Similarly, Xiaohongshu opened automatically, the search box was filled, and results appeared.
My thoughts:
This method is indeed the simplest; no technical skills are needed, just an internet connection.
However, it has clear drawbacks:
The cloud phone isn’t your real phone, so it can’t access your local apps and data, and its functions are relatively basic.
It’s suitable for experience but not for daily use.
Method 2: Local Deployment (30 Minutes)
Target Audience: Those with some computer knowledge who want to use it on a real phone.
This method requires connecting your phone to your computer, allowing the AI to operate your actual device.
My Process:
Preparation:
Before starting, I prepared the following:
- An Android phone (I used a Xiaomi 11)
- A computer (Windows 10)
- A USB data cable
- About 30 minutes of free time
Step 1: Install ADB Keyboard on your phone
This is a crucial step, as AutoGLM uses it to control your phone.
Download ADB Keyboard from GitHub:
https://github.com/senzhk/ADBKeyBoard
Find the latest APK file, download it to your phone.
Install the app:
Open the downloaded APK on your phone and install it.
You may see a prompt saying “Installation from unknown sources is not allowed”; just allow it in settings.
Set it as the default input method:
This step is important:
Open your phone’s “Settings” → Find “Language and Input” (location may vary by phone) → Find “ADB Keyboard,” click to enable it and set it as the default input method.
After setting, the keyboard you see when typing will be ADB Keyboard.
Step 2: Install ADB tools on your computer
ADB stands for Android Debug Bridge, which allows communication between your computer and phone.
Windows user steps:
Download the ADB tool package:
I downloaded the official Platform Tools:
https://developer.android.com/studio/releases/platform-tools
Choose the Windows version, download, and extract it to any directory.
I extracted it to D:\adb-tools.
Add to environment variables:
If you don’t do this step, you’ll need to switch to that directory every time you use ADB, which is cumbersome.
Steps:
Right-click “This PC” → “Properties” → “Advanced system settings” → “Environment Variables” → In “System Variables,” find “Path,” click “Edit” → Click “New,” enter the ADB tool path, e.g., D:\adb-tools → Click “OK” to save.
Verify installation:
Open CMD (Command Prompt) and enter:
adb version
If the version number displays, the installation was successful.
I sighed in relief when I saw this message:
Android Debug Bridge version 34.0.5
Step 3: Connect your phone and computer
This is the easiest place to encounter problems; I hit several snags.
Phone settings:
Enable Developer Options:
The path may vary by phone, but it’s generally similar:
Open “Settings” → “About Phone” → Find “MIUI Version” or “Build Number” (tap quickly 7 times) → You’ll see a prompt saying “You are now in Developer Mode.”
Enable USB Debugging:
Return to “Settings” → “More Settings” → “Developer Options” → Find “USB Debugging” and turn it on.
Allow USB debugging:
After connecting your phone to the computer with a USB cable, a prompt will appear on your phone:
“Allow USB debugging?”
Click “Allow.”
Important: Remember to check “Always allow this computer to debug,” or you’ll have to click allow each time.
Verify connection on computer:
In CMD, enter:
adb devices
If everything is normal, you’ll see output like:
List of devices attached abc123def456 device
“device” indicates a successful connection.
The first time I didn’t succeed; it showed:
List of devices attached (empty)
Troubleshooting process:
I tried these methods:
- Change USB ports
- Switched from a front USB 2.0 port to a rear USB 3.0 port, and it worked.
- It turned out it wasn’t the cable issue; it was the port’s insufficient power supply.
- Restart ADB service
- In CMD, execute:
adb kill-server
adb start-server
Then reconnect the phone.
- Reinstall phone drivers
- Some phones require official drivers to be recognized by the computer.
- I downloaded Xiaomi’s USB driver from the official website, and it worked fine afterward.
Step 4: Download and install AutoGLM
Once connected successfully, you can install AutoGLM.
- Install Python environment
AutoGLM requires Python 3.8 or higher.
First, check if you have Python:
python --version
If not installed, download it from:
https://www.python.org/downloads/
Make sure to check “Add Python to PATH” during installation.
- Download AutoGLM
In CMD, navigate to the directory where you want to install it, for example:
cd D:\projects
Then execute:
git clone https://github.com/zai-org/Open-AutoGLM.git
If you don’t have git, you can download the ZIP file from GitHub and extract it.
- Install dependencies
Navigate to the AutoGLM directory:
cd Open-AutoGLM
Install the dependencies:
pip install -r requirements.txt
This step might take a while as there are many packages to install.
I initially failed to install, receiving a timeout error for a package.
Solution:
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
Using Tsinghua’s mirror source speeds things up significantly.
Step 5: Run AutoGLM
Everything is ready, and it’s finally time to run it!
First run:
In CMD, execute:
python run_autoglm.py
The first run will automatically download the model, which takes some time.
I waited about 20 minutes for the model to finish downloading.
The downloaded files are quite large, around several GB.
If the network is poor, you can manually download the model files and place them in the specified directory.
Success indication:
When you see this message, it means AutoGLM has started successfully:
======================================== AutoGLM is ready!
Model loaded: GLM-4-9B Vision model loaded Ready to accept commands…
Step 6: Practical testing
The exciting moment has arrived!
I tried entering the first command:
“Help me search for milk tea on Meituan.”
Then, I watched the phone screen—
The phone screen lit up!
The Meituan app opened automatically, the search box appeared, and the words “milk tea” were automatically entered, displaying search results.
I truly witnessed my phone operating itself!
This feeling was incredible, as if an invisible person was helping me operate the phone.
I tested several scenarios:
Scenario 1: Send WeChat message
Command:
“Send a WeChat message to Zhang San: Are we having dinner together tonight?”
Operation process:
Open WeChat → Find Zhang San’s chat window → Click the input box → Enter “Are we having dinner together tonight?” → Click send.
Success!
Scenario 2: Order takeout
Comments
Discussion is powered by Giscus (GitHub Discussions). Add
repo,repoID,category, andcategoryIDunder[params.comments.giscus]inhugo.tomlusing the values from the Giscus setup tool.