Rapid advances in Large Language Models (LLMs) and broad potential applications in biological research make them a compelling point of investigation. Key challenges remain, including the tendency of LLMs to hallucinate if they are not provided with domain-specific information, thereby potentially misinforming users; the small number of LLM applications directed to prokaryotes despite significant advances in microbiome studies, especially when compared to the many human-specific and eukaryotic domain-specific LLM application that have recently been released; and the inability of current microbe-specific LLM tools to provide sufficiently comprehensive, accurate, and timely answers with proper citations.

We present MicroTraitLLM, a retrieval-augmented generation (RAG) LLM which utilizes zero-shot and single-shot prompting to give specific, citation-based answers for researchers. Its connection to the live-updating PubMed Central Open Access article database allows the tool to remain up-to-date on scientific knowledge. MicroTraitLLM flexibly allows the user to customize their experience by selecting their choice of LLMs. The tool is also able to generate accurate citations in various formats. We present empirical results demonstrating that MicroTraitLLM provides both improvements in relevant literature search and informative responses as judged by microbial experts, while not increasing latency time compared to popular commercial LLMs.