s3geo is an MIT-licensed, plugin-based GeoAI backend with a React workbench. A question is turned into a structured query spec, planned as a DAG of spatial operations and executed by plugins.
- Validated before execution — every LLM-generated plan is checked first, and every plugin run keeps an execution trace.
- Deterministic + LLM paths side by side, so plans can be compared.
- English and Persian questions.
- MCP server exposing each capability as a tool for AI agents.
- 100 capabilities across vector overlay, spatial statistics, decision analysis (AHP, TOPSIS), raster and terrain, network routing, charts and data connectors.
- Data from raster/vector uploads, CSV, WMS, WFS, PostGIS and remote URLs.
- 1,800+ automated tests in CI.
pip install "smart-spatial-system[raster,pdf]"
smart-spatial-api serve --port 8000
In a case study on health-care access in Istanbul, the system reproduced the expert result in 20 of 20 unhinted LLM runs (Jaccard 1.0); a Cairo raster pipeline became ~20× faster (18.5 min → 55 s) with identical results.