android-use is a tool designed for natural language-driven automation on Android devices. It enables automated operations on Android phones using natural language commands.
- Precise Positioning & Interaction: Utilizes XML parsing to highlight interactive elements and supports clicking via index, significantly enhancing the Agent's positioning capabilities.
- Broad Model Support: Supports domestic models like Deepseek, Kimi-k2, GLM, etc. It works effectively even without vision capabilities by leveraging XML parsing.
Before using this tool, ensure USB debugging is enabled on your Android device.
- Device Requirements: Android 7.0+.
- Enable Developer Mode:
- Go to Settings -> About Phone.
- Find Build Number and tap it continuously about 10 times.
- Wait for the "You are now a developer!" toast message.
- Note: Steps may vary slightly by phone manufacturer. Search online if you can't find it.
- Enable USB Debugging:
- Go to Settings -> Developer Options.
- Find and enable USB Debugging.
Install the uv package manager from the official website.
MacOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | shWindows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"Option 1: WebUI (Recommended)
uvx android-use webui1224.mp4
Option 2: CLI (Interactive)
uvx android-useIf you want to contribute or develop locally:
git clone https://github.com/languse-ai/android-use
cd android-use
uv sync
# Start CLI
python -m android_use.cli
# Start WebUI
python -m android_use.appHere are three demos showcasing the capabilities:
Task: Open WeChat, search for "Doubao Phone" in Video Accounts, get all comments from the most liked video on the current page, share this video with a friend, and send a summary of the comments.
1222-1.mp4
Task: Open Douyin, search for "Anta Basketball Shoes" in the shop, and retrieve information for the top 10 products.
A0F23DBA-AF73-4B2F-88B9-6FA893B17107.mov
Task: Open Xiaohongshu, search for "browser-use", select the most liked post on the current page, like it, leave a concise, human-like comment, and summarize the core content of the post.