Mano - A GUI intelligent operation model launched by Minglue Technology

What is Mano?

Mano is a proprietary large-scale model developed by Mininglamp Technology, focusing on intelligent operation of graphical user interfaces (GUIs). Based on a multimodal foundational model, the model achieves state-of-the-art (SOTA) results in both Mind2Web and OSWorld benchmark tests through innovative technologies such as online reinforcement learning and automatic training data collection. Mano can accurately identify and manipulate GUI elements in web and desktop environments to complete complex tasks, such as filling out forms and logging into accounts, providing efficient solutions for automated operations and driving the development of the GUI intelligent agent field.

Mano's main functions

• Automated web page operationsThe model can automatically complete various operations on a webpage, such as filling out forms, clicking buttons, entering text, and submitting forms. It can be applied to scenarios such as automated data collection and automated webpage testing.
• Desktop application operationIt supports operations on desktop software, including opening the software, making menu operations, entering text, and clicking buttons.
• Cross-platform operationIt is compatible with multiple operating systems and browsers, enabling automated operations on different platforms and meeting diverse automation needs.
• Data collection and analysisIt supports automatically collecting data from web pages or desktop applications, performing preliminary analysis, and providing support for subsequent data processing and decision-making.
• Error Detection and RecoveryIt has an error detection mechanism that can promptly detect errors during operation, attempt to automatically recover, and improve the reliability and stability of operation.

Mano's technical principles

• Multimodal basic modelBased on a multimodal model, it can understand and process visual information (such as webpage screenshots) and text information (such as user commands and webpage text), enabling perception and understanding of the GUI environment.
• Online reinforcement learningThrough online reinforcement learning, Mano can continuously learn and optimize its operational strategies through interaction with the real environment, thereby improving its adaptability and decision-making ability in dynamic environments.
• Automatic collection of training dataDesign an automatic training data acquisition module to automatically generate and collect high-quality interactive data for model training and optimization, reducing the cost of manual annotation.
• Supervisory fine-tuning (SFT)In the first stage of training, supervised learning is used to fine-tune the model, which can better understand and perform specific GUI operation tasks.
• Offline reinforcement learningIn the second stage, offline reinforcement learning is used to further optimize the model's decision-making ability, enabling it to better complete multi-step operation tasks.
• Online reinforcement learningIn the third stage, the model interacts with the real environment in the simulated environment, and further improves the model's adaptability and flexibility through online reinforcement learning.
• Verification moduleMano is equipped with a verification module to verify the correctness of each operation, promptly detect and correct errors, and improve the accuracy and reliability of operations.

Mano's project address

• Technical PapersLink: https://www.mininglamp.com/wp-content/uploads/2025/09/%E6%98%8E%E7%95%A5%E7%A7%91%E6%8A%80-Mano-Technical-Report.pdf

Application scenarios of Mano

• Automated data acquisitionMano can automatically collect data from web or desktop applications, providing efficient support for data analysis and research and saving manual collection time.
• Web page automated testingThe model automatically completes various operations on a webpage, such as filling out forms and clicking buttons, and is used to test whether the webpage functions properly, thereby improving testing efficiency and accuracy.
• Enterprise Automated OfficeIt supports the operation of desktop software, enabling the automation of internal office processes within enterprises, such as automatically filling in reports and sending emails, thereby improving work efficiency.
• Software Automated TestingThe model is used to automate the testing of desktop software, including opening the software and performing menu operations, helping developers quickly discover software problems.
• Intelligent customer service assistanceThe model can automatically handle some common customer inquiries, such as checking order status and answering frequently asked questions, reducing the workload of customer service staff.