GISclaw: An Open-Source LLM-Powered Agent System for Full-Stack Geospatial Analysis

A research team led by Jinzhen Han has developed GISclaw, an open-source agent system that bridges the gap between large language models and professional geospatial analysis. Unlike previous vector-only solutions tied to proprietary platforms, GISclaw offers full-stack capabilities across vector, raster, and tabular data types. The system integrates an LLM reasoning core with a persistent Python sandbox and comprehensive open-source GIS libraries including GeoPandas, rasterio, and scikit-learn. This architecture enables automated execution of complex workflows like spatial joins, terrain analysis, and predictive modeling.

GISclaw implements two pluggable agent architectures—a Single Agent ReAct loop and a Dual Agent Plan-Execute-Replan pipeline—and supports six heterogeneous LLM backends ranging from cloud-hosted models like GPT-5.4 to locally deployed 14B parameter models on consumer GPUs. Through three key engineering innovations—Schema Analysis for bridging task-data gaps, Domain Knowledge injection for specialized workflows, and an Error Memory mechanism for intelligent self-correction—the system achieved 96% task success on the 50-task GeoAnalystBench benchmark. Systematic evaluation across 600 model-architecture-task combinations revealed that the Dual Agent architecture consistently degrades strong models while providing marginal gains for weaker ones.

The researchers also proposed a novel three-layer evaluation protocol incorporating code structure analysis, reasoning process assessment, and type-specific output verification for comprehensive GIS agent assessment. This represents a significant advancement in how AI systems for specialized domains are benchmarked. All code and evaluation materials are publicly available, making GISclaw the first fully open-source solution of its kind that doesn't require expensive proprietary GIS software licenses while delivering professional-grade analysis capabilities.