The Electro-optical Targeting System (EOTS) performs detection, tracking, and identification of maritime and aerial targets using EO/IR sensors. However, in maritime surveillance environments, operators must manually configure workflows such as multisensor control and detection or tracking parameter settings, which requires considerable time and manpower and hinders rapid response. To overcome these limitations, in this study, we propose an EOTS-AI Agent framework based on a large language model (LLM). The proposed framework interprets and analyzes natural language queries from maritime surveillance operators, autonomously selects appropriate tools, and executes tasks step by step while providing an intuitive and efficient natural language interface. The results demonstrate that the proposed framework reduces operator workload and enables rapid, multicondition detection and tracking, thereby significantly improving the efficiency and reliability of maritime surveillance operations.