Large Language Models (LLMs) can produce human-like responses across various domains from healthcare to history. Millions of people use those models and chatbots daily, and some use them to get responses in critical domains like healthcare, law, and so on. However, the most concerning problem found in those models is hallucination, where false, misleading, or exaggerated information is produced. Many people use the models without cross-checking the content generated by the models, which can lead to some unforeseen consequences. Guaranteeing the trustworthiness and reliability of AI systems needs to be done with the utmost caution, given the sensitive nature of some domains. In this study, we use one of the most widely used methods in recent months to detect hallucinations: an LLM-based metric. We propose a novel and unique architecture based on recent advances in the LLM field, especially the advances in agentic AI. We call it “LieToMe”, a multi-agent-based system that we believe can detect hallucinations more accurately than existing systems. It includes 3 worker agents: a Retrieval Augmented Generation agent, a Web agent, and a Domain-based agent. In this paper, the focus is on the healthcare domain, although the modularity of the system allows it to adapt to other domains like law, science, and so on. The system computes a Composite Hallucination Score (CHS) that contains weights we can vary depending on the domain. We use some state-of-the-art models to test our hallucination detection system. Good results have been achieved, constituting solid proof of concept of this method.