UMLBot: Generative AI for converting natural language and code excerpts to editable UML diagrams

Unified Modeling Language (UML) provides a standardized visual representation that facilitates communication and comprehension of design ideas, particularly for software systems with complex architectures and behaviors. However, creating and maintaining UML diagrams is time-consuming and often requires knowledge of specialized syntax. We present UMLBot, an open-source web application that uses large language models to generate editable PlantUML source and rendered UML diagrams from natural-language descriptions or directly from code excerpts. UMLBot supports iterative refinement through a chat-based interface, allowing users to request changes, inspect the generated PlantUML, and regenerate diagrams as requirements evolve. The application includes configurable OpenAI-compatible LLM access, PlantUML rendering, self-correction and retry with flexible deployment through Docker Compose.