What it does

AutoGLM-GUI is a web-based control panel for AutoGLM, an AI agent that manipulates Android devices through ADB. You chat with it, it taps and swipes on your phone. The GUI adds scrcpy-based screen mirroring, multi-device management, cron-style scheduled tasks, and Docker packaging so you can park the whole thing on a server and let it run overnight sign-ins or periodic checks.

The interesting bit

The “layered agent” mode splits planning and execution across two models: a decision layer reasons about the task, then dispatches atomic sub-tasks to a vision-and-action layer. You can watch the tool calls unfold in the UI, which makes debugging opaque AI behavior slightly less like shouting into a void. It also exposes an MCP server, so Claude Desktop or Cursor can treat your phone as another tool.

Key highlights

• Scheduled tasks: Cron-style automation for repetitive phone chores (sign-ins, checks).
• Headless-friendly: Docker images for x64 and ARM64, meant for 24/7 server deployment.
• Wireless pairing: Android 11+ devices connect via QR code scan; no USB cable needed after initial setup on older versions.
• Multi-device: Control several phones or emulators from one dashboard with isolated state per device.
• Desktop builds: Pre-built EXE, DMG, AppImage, and deb packages for users who’d rather not touch Python.

Caveats

• The QR-pairing feature needs mDNS multicast, which the README warns may break in Docker bridge networks; --network host is strongly recommended.
• Requires an OpenAI-compatible API endpoint (BigModel, ModelScope, or self-hosted vLLM/SGLang); the model itself is AutoGLM-Phone-9B.
• Auto-update for portable builds (Windows EXE, Linux tar.gz) is explicitly not supported.

Verdict

Worth a look if you need to automate Android apps at scale or run phone-based tasks unattended. Skip it if you don’t already have an OpenAI-compatible model endpoint handy or if your Android fleet is stuck below version 11 and you hate cables.