GeoAgent: An Agent-Based QGIS Plugin for Natural-Language-Driven Geospatial Analysis with Multi-Backend Large Language Model Support

Tek Bahadur Kshetri , Rabin Ojha

Geospatial analysis is essential across the earth and environmental sciences, yet conventional geographic information systems (GIS) rely on a procedure-oriented paradigm that demands substantial operational expertise. We present GeoAgent, an open-source QGIS plugin that enables goal-driven geospatial workflows through natural language conversation powered by large language models (LLMs). GeoAgent features two operating modes: a General mode for exploratory querying and layer management, and a Processing mode for multi-step workflow execution. Built on LangGraph state machines, the system decomposes requests into dependency-ordered task queues, discovers algorithms across the entire QGIS Processing registry (~250+ native, GDAL, GRASS GIS, and SAGA tools) via lexical scoring on LLM-generated keywords, introspects parameter schemas at runtime, and executes workflows with LLM-steered error diagnosis and retry loops. Schema-validated structured outputs ensure consistent performance across interchangeable cloud (OpenAI, Google Gemini, Anthropic) and local privacy-preserving (Ollama) backends. We demonstrate GeoAgent through a reproducible flood-exposure screening case study using a bundled dataset and compare its architecture against existing LLM-GIS integrations. GeoAgent is released under the MIT license on the official QGIS Plugin Repository.