How to Run JackDaw Locally

The PoliRuralPlus project concluded its 2025 webinar series with “How to Run JackDaw Locally,” held on December 19th at 13:00 CET. This session focused on the JackDaw tool, an AI-powered chatbot for geospatial analysis, and explored its technical implementation, philosophical roots, and future development directions. The discussion was led by Karel Charvat (Czech Center for Science and Society), one of the project’s key developers, who offered both a historical and forward-looking perspective on the intersection of AI and Geographic Information Systems (GIS).

Setting the Stage: From Concept to Practice

Karel opened the session with a reflection on the broader evolution of AI in GIS, recalling early dialogues around whether “AI would eat Geo.” He argued that while AI may not replace traditional GIS, it is fundamentally transforming how spatial data is accessed, analyzed, and understood. Through this transformation, JackDaw aims to democratize geospatial analysis, enabling non-experts to interpret complex spatial data while preserving the critical role of geodata professionals in ensuring data quality and semantic precision.

Integrating Data and Intelligence

The webinar offered a deep technical overview of JackDaw’s architecture. Participants explored the integration of raster, vector, and unstructured data, alongside the use of RAG (Retrieval-Augmented Generation) and GeoRAG technologies. These innovations allow JackDaw to retrieve, interpret, and generate geographically grounded answers using spatially indexed knowledge bases.

Karel emphasized the importance of semantic metadata, moving beyond traditional standards like INSPIRE and Stack to richer, AI-readable forms. Enhanced metadata supports automated reasoning, interoperability, and real-time analysis, paving the way for a new generation of intelligent geospatial assistants.

Local vs. Cloud Models: Finding the Balance

A central part of the discussion compared local and cloud-based AI deployments, focusing on trade-offs in cost, autonomy, performance, and data sovereignty. Live demonstrations showed how locally hosted models, including those on Cesnet HPC infrastructure, can deliver competitive results while offering greater control and security.

The team also presented results from experiments using GeoRAG with the LAMAS 3.2 model, producing structured outputs (e.g., JSON) suitable for integration with other analytical tools, a step toward machine-to-machine geospatial reasoning.

Future Collaboration and Code Camp Invitation

The session closed with a look ahead to continued experimentation and community engagement. Karel invited participants to join the upcoming Prague Code Camp in January 2026, an open forum for co-developing new tools, models, and metadata standards supporting JackDaw and related PoliRuralPlus technologies.

“We are not just building another GIS tool — we are redefining how people and machines think about space, data, and meaning.”
- Karel Charvát, CCSS