This paper presents an innovative approach to software code review using Large Language Models (LLMs), incorporating open-source models. We introduce Pearbot, an open-source tool that implements a comprehensive code review workflow using open-weights LLMs, featuring multi-agent capabilities and reflection mechanisms. Our approach demonstrates the potential for LLMs to identify code issues and suggest improvements that may be overlooked by traditional automated analysis tools, while addressing common limitations of LLM-based code review through a multi-agent architecture.
Rybalchenko et al. (Wed,) studied this question.