Geliştiriciler DeepSeek aracılığıyla bilgisayarı yönetmek için bir araç sundu

Enthusiasts launched the dsh-computer-use_codex-style project, which implements desktop management functionality in the style of Codex based on the DeepSeek Harness infrastructure. The tool allows AI agents to interact with the Windows operating system and the Chromium browser, performing actions directly in the application interface.

The system supports 13 specific control methods, including commands for mouse clicks, text input, and application launch. Interaction occurs through real-time analysis of desktop screenshots, allowing the model to interpret the visual context and make decisions about further steps without the need for API integrations of specific programs.

To work with the solution, standard input methods are used that mimic user actions. The project is focused on automating routine tasks in a desktop environment, providing full control over active windows and browser tabs. The project code is available in an open repository for integration with local models.

Implementing this approach allows expanding the capabilities of LLM beyond text chats, turning them into full-fledged agents for working with graphical interfaces. Using DeepSeek Harness as a basis ensures compatibility with current methods of visual data processing for computer vision and automation tasks.