SuperAPI AI Directory
Partner with us中文AI newsFavorites
DirectoryNews and tutorials

AutoGLM - Zhipu AI's world's first phone Agent

Author: AI小集 · Original source: AI工具集

What is AutoGLM

AutoGLM is the world's first phone Agent launched by Zhipu, providing users with efficient operation agent services through AI technology. AutoGLM is based on domesticGLM-4.5andGLM-4.5Vmodels, with reasoning, coding, and multimodal capabilities. AutoGLM implements cloud-based execution without occupying the user's local device resources. Users can use voice commands to have it complete various operations, such as ordering takeout, booking flights, and checking housing listings. AutoGLM supports cross-device operations, breaking through hardware limitations. AutoGLM's operation execution capabilities are encapsulated as an API, allowing developers to integrate them into various hardware devices, such as AI glasses, watches, and home appliances.

On March 31, 2025, Zhipu launched the first free AI Agent with deep research and operation capabilities:AutoGLM Thinking

AutoGLM

Main features of AutoGLM

  • Cloud-based operation agent：
    • Life scenarios: Users can use voice commands to have AutoGLM operate dozens of high-frequency apps such as Meituan, JD.com, Xiaohongshu, and Douyin to complete tasks such as ordering takeout, booking flights, and checking housing listings.
    • Office scenarios: AutoGLM can execute full-process work across websites, operating web versions of Feishu, NetEase Mail, Zhihu, Weibo, Douyin, Micro Headlines, and other websites to complete information retrieval, content writing, video generation, PPT or podcast creation, and directly publish to social media platforms.
  • Independent operation：
    • Dedicated agent phone/computer: AutoGLM equips AI with dedicated cloud-based devices, enabling it to run independently without occupying the user's local device resources, so users can use other apps at the same time.
    • Asynchronous agent office work: Users can have AutoGLM continue executing tasks while offline, achieving 24-hour uninterrupted service.
  • Cross-device operation：
    • Breaking through hardware limitations: AutoGLM can run on any device and in any scenario, supporting cross-device operations such as phones, computers, watches, glasses, and home appliances.
    • API encapsulation: AutoGLM's operation execution capabilities are encapsulated as an API, allowing developers to integrate them into various hardware devices, giving hardware complete phone-level operation capabilities..

Technical principles of AutoGLM

  • Driven by domestic models: AutoGLM is based on Zhipu's latest open-source SOTA language model GLM-4.5 and visual reasoning model GLM-4.5V, with reasoning, coding, and multimodal capabilities. The models can complete multiple types of tasks such as reasoning, coding, research, Agentic and GUI operations, and flexibly call the most suitable "brain" to complete execution according to needs.
  • End-to-end asynchronous reinforcement learning：
    • ComputerRL: Proposes the API-GUI collaborative paradigm to improve data diversity and computational efficiency; improves GRPO and proposes the Entropulse mechanism to enhance exploration and policy diversity.
    • MobileRL: Innovates a difficulty-adaptive reinforcement learning method (reasoning bootstrap warm-up + difficulty-adaptive GRPO), significantly improving the stability and convergence efficiency of mobile tasks.
    • AgentRL: Through cross-sampling and task advantage normalization mechanisms, it solves instability and uneven gradient distribution in multi-task training, enhancing overall robustness and efficiency.

How to use AutoGLM

  • Online experience：
    • Visit the official website: Visit the AutoGLM official website https://autoglm.zhipuai.cn/。
    • Start the cloud device: After the page finishes loading, click the start button to start AutoGLM's cloud device.
    • Interact with AutoGLM: Enter a question or instruction in the input box.
      • For example, enter the following instruction:
      • "Help me check flights from Beijing to Shanghai for tomorrow."
      • "Help me order a cup of coffee."
      • "Help me generate a PPT about environmental protection."
    • View results: AutoGLM processes the request in the cloud and returns the results to the page.
  • Client：
    • Downloadable app: Download the AutoGLM client.
    • Registration and login: Complete account registration and login to enter the app.
    • Voice commands: After opening the app, tap the voice input button and speak the command. AutoGLM automatically recognizes the voice command and executes the corresponding operation in the cloud.
    • Text input: Enter a text command in the app's input box. After entering it, tap the "Send" button, and AutoGLM will perform operations based on the command.
    • View results: After AutoGLM completes the task in the cloud, it returns the results to the device.

AutoGLM project address

  • Official website address：https://autoglm.zhipuai.cn/
  • Project official website：https://xiao9905.github.io/AutoGLM/

AutoGLM application scenarios

  • Social media management: Automatically perform operations such as liking, commenting, and sharing on social platforms.
  • Online shopping: Search for products, compare prices, place orders, track logistics, etc. on e-commerce platforms.
  • Travel booking: Search for and book hotels, flights, train tickets, etc. on travel websites.
  • Food delivery ordering: Browse menus, place orders, make payments, and track order status on food delivery platforms.
  • Daily information queries: Such as checking weather, news, stock information, etc.