This paper discusses the application of large language models in a system based on the use of LLM-agents to automate business processes on a smart farm. The hypothesis is formulated that the information system in combination with computer vision models and machine learning algorithms is capable of solving business processes existing on a smart farm. The main focus is on the development of two types of LLM agents: a yield calculation agent and a Self-RAG based Q/A agent. The main components of the agent and the sequence of request processing are planned. Agents using LangChain and LangGraph were developed. System and agent architectures were designed. Integration of agents into a single architecture was carried out to perform various tasks such as counting and classifying the strawberry crop and calculating the required number of containers, and generating answers to user questions based on a knowledge base. The counting agent applies the YOLOv8 computer vision model to identify and count ripe berries, then calculates the approximate weight of the crop and the required number of containers for its storage and transport. A Q/A agent based on the Self-RAG approach is implemented, it uses a vector embedding model and a search engine to provide relevant answers to questions posed by users. An LLM-based router is implemented for query routing, it determines to which agent to redirect the query based on the given instructions. The classification of router queries is compiled. The results of evaluating the accuracy of the router using different LLM models as well as the prompt used are presented. Routing examples and examples of responses to different types of queries are given. Quality assessment comparison of response generation using different approaches has been made. The advantages of using LLM-agents on a smart farm, the prospects for their future use and possible improvements to the architecture are discussed.