Agents / AutoGLM

AutoGLM

Zhipu AI's open-source phone-use autonomous agent (repo Open-AutoGLM). The VLM sees the phone screen, plans a chain-of-thought action sequence, and executes it via ADB (Android), HDC (HarmonyOS NEXT) or WebDriverAgent (iOS). First phone agent with true Phone Use capabilities (2024-10-25); AutoGLM 2.0 commercial product runs agents in cloud virtual phones. Research/learning use.

AutoGLM is a autonomous agent by zhipu-ai, deployable as both a cloud service and self-hosted, and released open source.

Last verified: 2026-09-20 · Data status: Current · Next review: 2026-10-20

Key facts for AutoGLM
Key fact	Value
Company	zhipu-ai
Type	autonomous
Framework	Python framework (PhoneAgent API + CLI); model AutoGLM-Phone-9B built on the GLM-4.1V-9B family
Deployment	both
Open source	Yes (Apache-2.0 (code); MIT (models))
GitHub	https://github.com/zai-org/Open-AutoGLM
Documentation	https://docs.bigmodel.cn/cn/guide/models/vlm/autoglm-phone.md
Pricing	Open-source framework free. AutoGLM-Phone API (BigModel model id autoglm-phone) limited-time free as of 2026-09-20; paid price after promotion not publicly disclosed.

Capabilities

Capabilities of AutoGLM
Capability	Supported
Tool calling	Yes
Browser use	Unknown
Computer use	Yes
MCP support	Unknown
Memory	No
Planning	Yes
Multi-agent	Unknown
API	Yes

Use cases

Food delivery ordering and reordering
Product purchase and cross-platform price comparison
Travel planning (routes, flights/trains, hotels)
News, music, video playback and social interactions
Cloud-phone batch operations - notifications, likes, customer-service and attendance workflows
AI-native phone research and GUI agent development

Limitations

Research/learning use only; prohibited for illegal information gathering
Sensitive pages (payment, password, banking) cannot be screenshotted - agent auto-detects and requests human takeover
Android 7.0+ with developer mode + USB debugging required; ADB Keyboard needed for Android text input
iOS support requires separate WebDriverAgent setup
Local deployment needs ~24GB+ VRAM GPU
No persistent-memory feature documented in the README

AutoGLM is Zhipu AI’s open-source phone-use agent (Open-AutoGLM): the vision-language model reads the phone screen, plans a chain-of-thought action sequence, and executes it over ADB (Android), HDC (HarmonyOS NEXT) or WebDriverAgent (iOS). It was the first phone agent with true Phone Use capabilities (2024-10-25); the commercial AutoGLM 2.0 product runs the same approach on cloud virtual phones.

The framework ships as a Python package (PhoneAgent API + CLI) built around AutoGLM-Phone-9B, a 9B VLM from the GLM-4.1V-9B family. The open-source framework is free, and the hosted AutoGLM-Phone API on BigModel was free during its promotional period as of 2026-09-20.

Local deployment needs a GPU with roughly 24GB+ VRAM; Android use requires developer mode and USB debugging, and iOS needs a separate WebDriverAgent setup. See the Zhipu AI profile for the wider GLM ecosystem.

Sources

[Official] Open-AutoGLM GitHub repository [Official] AutoGLM Goes Open Source blog [Official] AutoGLM-Phone model card (Hugging Face) [Official] BigModel AutoGLM-Phone API docs

Data interpretation

Evidence types for AutoGLM
Field	Value	Evidence type
Company	zhipu-ai	Official
Type	autonomous	Official
Open source	Yes (Apache-2.0 (code); MIT (models))	Official
Deployment	both	Official
Capabilities	Tool calling, Computer use, Planning, API	Official

Evidence types: Official = vendor documentation or official pages. See the sourcing policy.

What is AutoGLM?

Is AutoGLM open source?

Yes — licensed under Apache-2.0 (code); MIT (models).

How much does AutoGLM cost?

Open-source framework free. AutoGLM-Phone API (BigModel model id autoglm-phone) limited-time free as of 2026-09-20; paid price after promotion not publicly disclosed.

Where does China AI Hub get its AutoGLM data?

From 4 sources, last verified 2026-09-20.