AppAgent is a Python-based multimodal agent framework that operates smartphone apps via an LLM-driven action space, with GPT-4V or Qwen-VL-Max as options. It provides a configuration-driven setup, a two-phase exploration/deployment workflow, and scripts to learn and run tasks on Android devices.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
AppAgent: Multimodal Agents as Smartphone Users, an LLM-based multimodal agent framework designed to operate smartphone apps.
How it works
The framework enables an agent to interact with mobile apps through actions like tapping and swiping, learning to navigate apps via autonomous exploration or human demonstrations, and producing documentation of interacted elements for deployment.
Getting started
"To configure your requests to GPT-4V, you should modify config.yaml in the root directory." The Quick Start section describes prerequisites, installation, and how to run the provided scripts. Commands shown include:
cd AppAgent
pip install -r requirements.txt
python learn.py
python run.py
Configuration notes mention two key parameters: OpenAI API key and Request interval. Alternative multi-modal model option: qwen-vl-max with a Dashscope API key filled into DASHSCOPE_API_KEY in config.yaml, and set MODEL from OpenAI to Qwen.
Recent releases
Latest release entry: none.
Traction
Open-source repository with 6839 stars, 758 forks, 99 open issues. Language: Python. License: MIT. Created: 2023-12-20. Last push: 2025-03-19.
Behind the repo
Linked project page references and a project team list in the README are present, including contributors and a project leader, but no separate startup/company link is provided in the FACTS section.
Caveats
License: MIT. Created: 2023-12-20. Last push: 2025-03-19. Open issues: 99. Language: Python. No explicit note about age beyond creation date in the facts.