Abstract for: From Running Models to Understanding Them: Conversational Interfaces for System Dynamics Dashboards

System dynamics models are increasingly deployed through web-based dashboards, expanding access to non-modelers. However, while dashboards make models easier to run, they do not necessarily make them easier to understand. This project addresses that gap by integrating a model-grounded conversational assistant into simulation dashboards to support real-time interpretation of model behavior during user interaction. We embed a conversational assistant within a web-based dashboard, enabling direct access to model inputs, outputs, and documentation. A structured model contract defines the information exposed to the assistant. At query time, the system constructs a bounded context using current simulation state and approved materials, allowing a GPT-based model to generate responses grounded in model structure and scenario-specific outputs. A working prototype has been implemented and tested across multiple model environments. The assistant can explain variables, interpret simulation outputs, and guide scenario exploration. Early testing shows that structuring inputs as explicit model state and trajectories improves response consistency. Users are able to ask follow-up questions and receive explanations that connect observed trends to feedback processes, though formal evaluation has not yet been conducted. This work introduces a conversational layer as a new interface for interacting with system dynamics models to support deeper engagement and interpretation. Ongoing work will evaluate its impact on user understanding and identify design principles for conversational support in simulation environments, with implications for model dissemination and decision support.