Graph of Goal-Oriented Thoughts: Design and Implementation of LLM Agents

This paper presents a novel framework for managing complex user interactions in conversational agents through a goal-oriented model that integrates both proactive and reactive querying of Large Language Models (LLMs). Drawing inspiration from goal-based agents and the "Graph of Thoughts" (GoT) paradigm, we propose a structured approach to goal management that tracks and reasons about goals and sub-goals. Goals are represented as explicit "thoughts" within the GoT framework, enabling transparent and observable reasoning processes. Through an illustrative example, the paper shows the flexibility and effectiveness of this approach in adapting to dynamic and complex contexts. Moreover, future developments could extend the applicability of this paradigm to more intricate domains.