suyiiyii/AutoGLM-GUI

AutoGLM‑GUI – AI‑powered Android automation workstation

What it is

• A desktop/web application that lets a large‑language‑model (any OpenAI‑compatible endpoint) control one or many Android devices (real phones or emulators) through ADB.
• It adds a chat‑style interface, task scheduling, workflow saving, and Docker/stand‑alone binaries so the assistant can run 24/7.

Key capabilities

Feature	What you get
Layered agent mode	Two‑stage architecture – a planning LLM breaks a request into steps, a execution visual model (AutoGLM‑Phone) performs the UI actions.
Cron‑style scheduled tasks	Define recurring jobs (e.g., daily sign‑in, price checks) that the AI will run automatically.
Multi‑device management	Connect several Android devices (USB or Wi‑Fi) at once; each device has its own chat history and screen preview.
Real‑time screen preview	Low‑latency video from the device via scrcpy; you can also click/drag directly on the preview to issue manual actions.
Docker & portable binaries	Pre‑built Docker image and Windows EXE / macOS DMG / Linux AppImage / deb / tar.gz for “run‑anywhere” deployment.
Conversation history	All chats, AI reasoning steps, screenshots and logs are stored locally (SQLite) and viewable in the UI.
MCP integration	Exposes a Model Context Protocol server so other AI tools (Claude Desktop, Cursor, Cline, etc.) can call list_devices() and chat(device_id, message) to automate phones from within those environments.
Instant interrupt	< 1 s to abort a running task, useful when the AI goes off‑track.
Zero‑config setup	The bundled desktop version ships with Python, ADB and required libraries; you only need to point it at an LLM endpoint and an API key.

How to get started

1. Desktop binary – download the appropriate package for Windows, macOS (Apple Silicon) or Linux and run it. The first launch asks for your LLM base URL, model name and API key (e.g., Zhipu BigModel, ModelScope, or a self‑hosted vLLM endpoint).
2. Python package – pip install autoglm-gui then autoglm-gui --base-url <url> --model <name> --apikey <key>.
3. Docker – docker run -d --network host -v autoglm_config:/root/.config/autoglm ghcr.io/suyiiyii/autoglm-gui:main (or use the provided docker‑compose.yml). The container runs the web UI on port 8000; open http://localhost:8000.

Typical workflows

• Personal assistant – type “order a cup of tea on Meituan” and watch the AI navigate the app, fill the form and place the order.
• Scheduled automation – create a cron job that every morning signs into a loyalty app, checks order status, and sends you a notification.
• Batch device management – connect several phones and assign each a different workflow (price monitoring, social‑media posting, game daily quests).
• CI/CD for mobile apps – run the emulator version, let the AI perform UI tests, capture screenshots and log results automatically.
• MCP‑driven tooling – from Claude Desktop you can call chat(device_id, "打开微信并发送消息…") to let the AI act on your phone without leaving the chat interface.

Why it matters AutoGLM‑GUI turns a generic LLM into a concrete Android‑automation agent, handling the heavy lifting of screen capture, coordinate conversion, and ADB communication. By bundling scheduling, multi‑device support, and an extensible MCP server, it bridges the gap between conversational AI and real‑world mobile actions, making it practical for personal productivity, small‑scale RPA, and mobile‑testing pipelines.

License – Apache 2.0 (the bundled ADB Keyboard APK is GPL‑2.0, isolated from the main code).