Building knowledge graph agents using LlamaIndex Workflows explores the integration of structured data into retrieval-augmented generation (RAG) frameworks, emphasizing the use of graph databases like Neo4j to enhance the accuracy and relevance of AI-generated responses. The article discusses the Text2Cypher approach, which translates natural language queries into Cypher statements, enabling the retrieval of relevant information from knowledge graphs. However, the technique faces challenges in precision due to language interpretation nuances. By employing LlamaIndex Workflows, the process introduces multi-step approaches, allowing retries and alternative query formulations to improve accuracy. The text also describes several architectures and workflows, including naive Text2Cypher, retry mechanisms, evaluation phases, and iterative planning systems, highlighting their strengths and areas for improvement. Benchmarking efforts reveal the effectiveness of certain models and workflows, emphasizing the importance of accuracy, stability, and speed in practical applications. The discussion concludes with insights on production challenges, such as handling real-world data and ensuring reliable system performance, while suggesting a focus on simple, effective implementations to build robust knowledge graph agents.