10/07/2026 –, Aula Magna
Current GIS stacks expose geographic data brilliantly — but no layer exists between raw geodata and agent cognition. AI agents querying PostGIS or running RAG on vector tiles are not geo-aware: they retrieve features, they don't perceive place.
GeoBard is an open service that addresses this gap. Given a GeoJSON viewport and contextual parameters — time of day, historical period, cultural domain — GeoBard performs semantic grounding: it synthesizes a situated natural language narrative of what an agent perceives at that location. Not a feature summary. A description of presence.
Built on top of OpenFantasyMaps and OpenHistoryMap structured geodata, GeoBard is source-agnostic and stack-composable: it sits between any OGC API / vector tile source and any agent runtime, with no LLM lock-in. It is the missing perceptual layer for geo-aware AI agents in FOSS infrastructure.
