HydroSuite-AI: a web-based LLM environment for hydrological code generation and execution for the hydrosuite open-source ecosystem

Abstract Researchers in hydrology face technical barriers from complex models, evolving libraries, and fragmented documentation. We present HydroSuite-AI, an LLM-based assistant that integrates the open-source web hydrologic libraries HydroLang, HydroCompute, and HydroRTC for code generation, factual guidance, and in-browser execution of hydrological workflows. The system uses a Planner-Worker-Synthesizer architecture with retrieval-augmented generation (RAG) over library documentation and validates outputs in a client-side virtualized execution environment. We evaluate HydroSuite-AI through (i) a cross-library case study, (ii) a controlled benchmark measuring the assistant’s ability to retrieve and execute library functions across 60 domain-specific tasks under different LLM backends, and (iii) deployment as a code assistant in the WaterSoftHack 2024 workshop. Across the benchmark tasks, HydroSuite-AI achieves 95.0% function retrieval accuracy with 23.1 s mean latency when powered by o3-mini, while the same system reaches 71.7% under 10s with GPT-4.1-nano, showing an accuracy–speed tradeoff across model configurations. The case study demonstrates end-to-end interoperability and exposes limitations in advanced statistical workflows, and the workshop deployment confirms the system’s utility in collaborative, time-constrained settings. Documentation-grounded LLM orchestration reduces time to implementation and supports hydrological research and education.