AppAgent is an open-source research framework by Tencent (QQ Games Research Lab) that enables large language model agents to operate smartphone applications by mimicking human interactions such as tapping, swiping, and text input. Agents autonomously explore apps or learn from human demonstrations, building a knowledge base to execute complex tasks. It was published at CHI 2025 and tested across 10+ apps and 50+ tasks including social media, image editing, and navigation. The project site is the GitHub Pages site at appagent-official.github.io. Key features: - LLM-based multimodal agent that controls Android apps via UI interactions - In-context learning from autonomous exploration or human demonstrations - Builds a persistent knowledge base per application - Supports complex multi-step tasks (e.g., posting on social media, editing photos) - Open-source Python codebase hosted on GitHub (TencentQQGYLab) - Tested on 50+ tasks across 10 diverse apps
Free, open-source (Apache 2.0).