Our default stack is open-source: QGIS for desktop analysis, PostGIS for spatial databases, GeoServer for serving map layers, and Leaflet or OpenLayers for the web maps clients actually interact with. This is a deliberate default, not a cost-cutting shortcut.
Why this is the default
Government and academic clients are especially cost-sensitive to recurring license fees, and open-source GIS tooling has matured to the point where it handles the vast majority of consulting and Web GIS product work without a functionality gap. It also avoids vendor lock-in — a department that owns its data in PostGIS and its map layers in GeoServer can move to a different developer or vendor later without starting over, which matters for a public-sector client thinking in decade-long timeframes, not quarterly budgets.
It also keeps our own margins predictable: we're not passing through per-seat license costs on every engagement, which is part of what lets us serve government and academic budgets that a license-heavy shop would price out of reach.
When we add commercial cloud AI on top
Open-source-first doesn't mean open-source-only. For specific AI/ML workloads — large-scale satellite image classification, some predictive risk-scoring models, or workloads that benefit from managed cloud infrastructure and GPU access — commercial cloud AI services can outperform what's practical to self-host, and we use them where the client outcome genuinely improves as a result.
The rule we apply: open-source is the default for data storage, serving, and standard analysis, because that's where vendor lock-in and license cost matter most over time. Commercial services get added selectively, on top of that open foundation, only where they add real analytical capability that open tooling can't yet match — never as a wholesale replacement for the base stack.