Open-AutoGLM: an open-source phone agent that can see your Android screen and execute tasks.

A practical, safety-first guide to evaluating Android GUI agents with reproducible setup steps and model options.

No 24GB GPU? Use a deployed model service endpoint if you have access.

Trust anchors from official sources

Trust anchors and official sources

Use official sources for code, models, and licensing. This site curates links and explains how to verify them.

Open source by default

Review the repository and run your own builds for reproducibility.

Official model sources

Download from Hugging Face or ModelScope (no re-hosting).

Clear licensing

Apache-2.0 terms for the code, with models governed by their own licenses.

Independent coverage

Not affiliated with the original authors; focused on safety and evaluation.

Two ways to run Open-AutoGLM

Option A is recommended for most users. Option B is for local power users.

1

Option A (recommended): model service endpoint

No local GPU required. Use a deployed model service endpoint and pass --base-url.

2

Option B (local): GPU deployment

Requires an NVIDIA GPU (24GB+ VRAM recommended), vLLM or SGLang, and ~20GB model files.

Why this site

An independent, educational resource for testing Android GUI agents safely.

Clear, practical docs

Step-by-step guidance for evaluating Open-AutoGLM without guesswork.

Transparent setup

Open-source, reproducible flows that surface hardware and safety tradeoffs.

Independent coverage

Not affiliated with the original authors; focused on objective evaluation.

Mobile Regression Testing (coming soon)

Join the waitlist for guided Android regression testing workflows and agent-safe checklists.

Support the independent Open-AutoGLM guide

Sponsor longform guides, deployment notes, or community spaces. Ads and Creem checkout are placeholders for now.

Open-AutoGLM: open-source Android phone agent