AI models handle most computer work well and maps badly. We measure how well they do GIS and Earth observation tasks, and we build spatial AI systems with teams on their own data and software. We publish what worked and what did not.
We have run 33 AI models, in 70 setups, on the same GIS and Earth observation tasks and checked every answer with code. Below are the top five in version 1.0.
A plugin for Claude and Codex. It rebuilds a legacy ArcGIS or QGIS project in Python, then compares every output with the original. A script decides whether they match, not the model.
Reeve reads a farm's records against the parcel register, aid rules, pest bulletins, prices and the forecast, and answers questions on Telegram. Every figure names its source. It runs on a sample farm with France's rules only.
Bring us a spatial problem your team works on every week. We build the AI system around your data and the software you already use, then run it with your team on real work until it holds up. What we learn goes into our research.
Listed as a new entrant in Geoawesome's 2026 Global Top 100 Geospatial Companies. See the list
