Retrieval-Augmented Generation (RAG) improves the reliability and capabilities of Large Language Models (LLMs) by integrating domain-specific knowledge, making it especially useful for simplifying complex geoscientific subsurface modeling software. Our research focuses on using RAG to build AI assistants that incorporate diverse contextual information—like documentation, test cases, and source code—to help both novice and advanced users navigate, troubleshoot, and optimize these modeling platforms. We demonstrate its effectiveness across various subsurface applications, including open-source and proprietary tools with different interfaces, showing that RAG-enhanced LLMs can boost productivity, ease learning curves, and improve software accessibility. Ultimately, this approach supports the democratization of sophisticated scientific tools in line with open science principles, fostering more inclusive and efficient geoscience research.