AutoGLM Phone 9B: Turn Your Android into an Autonomous AI Agent

Written by Matteo Giardino - CTO and Founder of OpenClaw.

AutoGLM Phone 9B is a breakthrough open-weights model capable of driving Android UI through autonomous interaction. If you've been waiting for a truly local, on-device AI agent capable of tapping, swiping, and typing just like a human, this is it. In this guide, I will show you exactly how to set up AutoGLM Phone 9B and connect it with OpenClaw.

Having built dozens of AI agents, I've noticed a missing piece: interacting natively with mobile applications without fragile Appium scripts. AutoGLM Phone 9B solves this by using multi-modal inputs (vision and text) to predict screen interactions directly.

What is AutoGLM Phone 9B?

AutoGLM Phone 9B is a 9-billion parameter multimodal language model optimized specifically for understanding and interacting with mobile graphical user interfaces. Unlike general-purpose LLMs, it was trained on vast amounts of Android screen recordings and UI layouts.

How It Works Under the Hood

The architecture relies on a continuous loop of screen observation and action prediction:

1. Perception: The model receives a screenshot of the current Android state.
2. Analysis: It identifies actionable UI elements (buttons, text fields, icons).
3. Action Prediction: Based on the user's natural language goal, it generates the exact coordinates and action type (tap, swipe, input).

Local Setup Guide

Running a 9B multimodal model requires decent hardware. You'll need at least a Mac with 16GB of Unified Memory or a PC with an RTX 3090/4090.

Here is how you initialize the agent:

# Clone the repository and set up the environment
git clone https://github.com/THUDM/AutoGLM.git
cd AutoGLM
pip install -r requirements.txt

# Run the local inference server
python server.py --model THUDM/AutoGLM-Phone-9B

Once the server is running, you can connect your Android device via ADB and execute the agent script to start automating tasks.

AutoGLM vs OpenClaw Ecosystem

How does this fit into the broader OpenClaw architecture? While PicoClaw handles lightweight API and desktop browser automation, AutoGLM bridges the gap for mobile-first environments. You can easily integrate AutoGLM as a specialized tool within your OpenClaw workflow.

Conclusion

Edge-deployed UI agents are the next frontier in automation. AutoGLM Phone 9B provides a robust, open-weights foundation for building Android automation without brittle selector-based scripts. It's time to build smarter workflows.