// the take
RustGIS is a desktop-and-web GIS written in pure Rust, aiming at the QGIS/ArcGIS niche: vector/raster editing, geoprocessing, cartography, and layout/print tooling in one app, plus CLI and Python bindings over the same engine. It's aimed at GIS analysts and devs who want a scriptable, cross-platform (native + WASM) alternative to QGIS without a Python/GDAL runtime to manage.
The crate layering (geom -> data -> map/analysis/tiles -> render/plugins -> engine -> assistant -> ui) is enforced by a CI layering check, not just a README diagram, which is unusual discipline for a project this young. It genuinely ships one engine three ways - native desktop, WebAssembly in-browser, and a Python package - rather than faking parity between them. The 'everything is a command' design (stable ids for every ribbon action, scriptable via CLI/ModelBuilder/notebooks) gives real reproducibility that QGIS's plugin API doesn't uniformly offer. Format coverage tracks the current geospatial stack (GeoParquet, FlatGeobuf, COPC LiDAR, PMTiles, DuckDB Spatial, STAC) instead of stopping at Shapefile/GeoTIFF.
The README states outright that the codebase 'is primarily built with Claude Opus 5.5... working under human direction and review' - fine for UI plumbing, but geodesic math, CRS/datum transforms, and kriging/interpolation are exactly the kind of code where subtle correctness bugs survive review and only surface after years of edge-case bug reports, which this project hasn't had time to accumulate. A large chunk of format/raster support ('every vector format GDAL reads when it is installed') is conditional on having GDAL installed, so the pure-Rust pitch doesn't actually cover full format parity on its own. The feature surface (geocoding, pixel classification, kriging, ModelBuilder, a plugin registry, an AI assistant, 1000+ geoprocessing tools) is enormous relative to a 2.2k-star, 332-fork community, which is a lot for a small maintainer base to keep correct and supported long-term. The AI Assistant forwards your data to whichever model you configure (Anthropic/OpenAI/Gemini/etc.) - fine for public data, but anyone with regulated or sensitive geospatial data needs to actually verify that's opt-in per session rather than assuming it from the README's phrasing.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →