AutoGLM

Chat

Z.ai

Product overview

Name of Agent: AutoGLM
Short description of agent: ["[...] AutoGLM, a new series in the ChatGLM family [11], designed to serve as foundation agents for autonomous control of digital devices through Graphical User Interfaces (GUIs)."](https://arxiv.org/html/2411.00820v1)
Date of release: 10/09/2025: AutoGLM 2.0 release 22/03/2025: AutoGLM 1.0 reflection release (link, archived) 28/10/2024: Initial release (link)
Advertised use: “Focusing on Web Browser and Phone as representative GUI scenarios, we have developed AutoGLM as a practical foundation agent system for real-world GUI interactions.” (link)
Monetisation/Usage price: Free (no information found on the web and iOS versions) accessed on 2025-12-09
Who is using it?: end users (web version, browser plugin and phone app users) API available upon request (link, archived)
Website: Web version (since Aug 2025): (https://autoglm.zhipuai.cn/, archived) iOS version (since Aug 2025): (https://apps.apple.com/gb/app/autoglm/id6748391183, archived) Initial release: (https://xiao9905.github.io/AutoGLM/, archived)
Category: Chat

Company & accountability

Developer: Z.ai
Name of legal entity: Beijing Knowledge Atlas Technology Joint Stock Company Limited 北京智谱华章科技股份有限公司 (link, archived)
Place of legal incorporation: Beijing, China
For profit company?: Yes
Parent company?: Not applicable
Governance documents analysis: AutoGLM Privacy Policy (2025 Jul 23): (https://autoglm.zhipuai.cn/privacy, archived) AutoGLM User Agreement (2025 Jul 8): (https://autoglm.zhipuai.cn/agreement, archived)
AI safety/trust framework: Developer Open Platform Self-Discipline Principles Statement (link, archived) Developer Open Platform Content Safety (link, archived)
Compliance with existing standards: Algorithm Registration. Zhipu ChatGLM Generation Algorithm (Network Information Registration No. 110108105858001230019). Zhipu ChatGLM Search Algorithm (Network Information Registration No. 110108105858004240011). Model Registration (link, archived) based on Interim Measures for the Management of Generative Artificial Intelligence Services (生成式人工智能服务管理暂行办法) (link, archived). 智谱清言（ChatGLM）：Beijing-ChatGLM-20230822.

Technical capabilities & system architecture

Model specifications: "Based on GLM-4.5, GLM-4.5V, with reasoning, coding, and multi-modality capabilities" (Blog post (link, archived))
Documention: (Release blog post, archived) (more technical details in ComputerRL, MobileRL, AgentRL - upcoming) Documentation not found on Developer’s Platform
Observation space: User input, Internet access, Computer GUI elements (Appendix C Prompt Formulation for AutoGLM-OS - Observation formulation from ComputerRL (link))
Action space: Sandboxed cloud computers GUI with Internet access (Detailed action space can be found on ComputerRL (link))
Memory architecture: Long-term memory for historical data and conversations (Appendix C Prompt Formulation for AutoGLM-OS - History formulation from ComputerRL (link)). No evidence of episodic memory across different conversations.
User interface and interaction design: Standard chatbot interface with an option to initiate web agents; Human users can take over the cloud computer GUI (e.g. to input passwords for account login purposes)
User roles: Operator (Users can run the agents) Executor (Users can interact with cloud computer GUI to provide human credentials and make decisions)
Component accessibility: Closed source

Autonomy & control

Autonomy level and planning depth: L4: User as an approver, user only interacts “when the agent encounters a blocker it cannot resolve on its own” such as a missing API key (based on example video demonstration of web agents (link, archived))
User approval requirements for different decision types: User input is needed for certain kinds of tasks (e.g., user authentication, payment, blog post release decisions)
Execution monitoring, traces, and transparency: Highly abbreviated action trace (such as listing out all searched webpage, clicking ), no visible CoT
Emergency stop and shut down mechanisms and user control: User can pause/stop the agent at any time
Usage monitoring and statistics and patterns: None found, AutoGLM is free to use on web version at the moment

Ecosystem interaction

Identify to humans?: Claims to have legally incorporated explicit and implicit identifiers into this product and its generated content, complying with the National Standard Measures for the Labelling of Artificial Intelligence-Generated and Synthetic Content (effective September 1, 2025) (link, archived). Nothing found on identifiers with non-user humans for AutoGLM. Images generated from AutoGLM (by calling self-developed image generation tools) has an AI-generated identifier at bottom-right, but could be irrelevant with web agents.
Identifies technically?: None found, CUA seems to interact with computer GUI like a human, without adding any specific identification to requests
Interoperability standards and integrations: None found, no mentions of AGNTCY, Agent Connect Protocol (ACP), Model Context Protocol (MCP), and Agent2Agent (A2A) protocol anywhere.
Web conduct: None found

Safety, evaluation & impact

Technical guardrails and safety measures: None found
Sandboxing and containment approaches: None found
What types of risks were evaluated?: None found, no mention of risk evaluation in ComputerRL (link), MobileRL (link) or AutoGLM Technical Report (link).
(Internal) safety evaluations and results: None found, no mention of risk evaluation in ComputerRL (link), MobileRL (link) or AutoGLM Technical Report (link).
Third-party testing, audits, and red-teaming: None found
Benchmark performance and demonstrated capabilities: Results for AIME 25, GPQA, LCB v6, HLE, and SWE-Bench Verified available for GLM-4.5 (link, archived). Computer use eval results available for AutoGLM (link). OSWorld, WebVoyager, AndroidLab, AndroidWorld (link, archived)
Bug bounty programmes and vulnerability disclosure: None found
Any known incidents or reported vulnerabilities?: None found
← Back to 2025 Index