Bridging Natural Language and Desktop GIS Automation with LLM-Powered GIS Plugins

Fehér Zsolt Zoltán

Geographic Information Systems (GIS) are indispensable for spatial analysis and remote sensing, but still, scripting interfaces that enable automation (ArcPy, PyQGIS, and SNAP GPT) impose a steep technical barrier on domain scientists who are not software developers. We present GIS~Chat, an open-source suite of three plugins that embed a Large Language Model (LLM) chat panel directly inside ArcGIS Pro, QGIS, and ESA SNAP Desktop. The user describes a desired operation in natural language; the plugin supplies five workspace contexts (open layers, bands, coordinate reference systems, selections) to the LLM, which generates and executes the corresponding platform-specific code through a tool-calling mechanism. As of version 1.1, GIS Chat integrates with Google Earth Engine (GEE), enabling users to query, process, and download satellite imagery and geospatial datasets directly from the chat panel without writing any GEE Python code. GIS Chat supports five LLM back-ends - including Google's Gemini and Ollama, whose adoption requires no paid subscription. By covering major desktop GIS/remote-sensing platforms with a single architectural pattern, GIS Chat lowers the entry barrier for non-programmers and thus aspires to provide a consistent conversational workflow regardless of the underlying software.