Task planning is essential in General Purpose Service Robot (GPSR), enabling robots to perform complex actions such as navigation, object manipulation, and human interaction. While traditional planning approaches rely on symbolic reasoning and deterministic rules, recent advancements in artificial intelligence (AI) have introduced Large Language Models (LLMs) as powerful tools for contextual reasoning and decision-making. This paper introduces GPSR_planning, a ROS 2-compatible component that utilizes LLMs for generating action plans from natural language commands. Designed for on-edge deployment, the system operates entirely onboard the robot, eliminating the need for cloud-based services and enhancing autonomy. The tool employs lightweight LLMs selected through benchmarking based on accuracy, inference time, and memory consumption. Evaluations conducted in simulated and real-world environments, including the RoboCup@Home setting, demonstrate the system’s effectiveness in dynamic and less-structured scenarios. The results highlight the potential of LLMs to improve interaction and adaptability in robotic planning tasks. The proposed approach marks a step toward more flexible, user-friendly, and deployable planning systems in service robotics.