Využití AI agenta pro správu linuxového systému

Abstract

This bachelor's thesis addresses the design and implementation of a desktop application that enables Linux system administration through natural-language requests processed by a large language model. The introductory part analyses selected existing tools and identifies their common shortcomings in the areas of safety and user interface design. Based on this analysis, an architecture is proposed that visually separates the conversational and working environments, supports multiple language-model providers and emphasises the protection of the user from incorrectly generated commands. The key safety mechanisms are separated user modes with different privilege levels, local validation of proposed commands on the application side and a file-system-level rollback mechanism. The functionality of the prototype is verified by a combination of unit tests and a set of functional scenarios covering typical request life-cycle situations including operations over system files. The conclusion summarises the comparison with the analysed tools, the identified limitations of the prototype and possible directions for further development.