Coding agents × QGIS Desktop
SpatialAgents are AI agents for geospatial data processing: they take a domain task and deliver analyses, maps and evidence directly in QGIS — with cloud models or entirely locally on your own network.
Codex CLI, Claude Code and OpenCode can each launch multiple SpatialAgents. They work on geospatial tasks in the shared workspace and use the technical bridge to access the open QGIS project, domain skills and the Geospatial API for AI agents.
This site shows the current state of research and development on the system. Projects, setup, training and support are provided by Geoinformatikbüro Dassau GmbH (GBD) ↗.
Locally proven · Benchmark 08 · Behördenakte
Seven agent-model systems solve the same hazard and climate analysis for Stuttgart in QGIS — among them Qwen3.8-Flash-Next on a notebook with a 70-watt GPU budget: 83.7 of 100 points, ahead of every GPT-5.6 cloud variant, with no request leaving your own network in local operation. In the Behördenakte dossier the same model reaches gold in 17 of 17 assessed runs on open ALKIS data.
Direct paths
Each route answers a distinct question, from the basic terms to the assessment of completed work.
00 / Fundamentals
Coding agent, tool use, MCP, skills, planning, bridge and measured evidence — each section with its own graphic, reading text and narration. The same sequence is available as a narrated presentation, 23 slides in 21 minutes.
01 / QGIS
The local bridge passes declared actions to QGIS and returns project state, layer information, screenshots and generated artefacts.
02 / Quickstart
The workflow leads from the plugin ZIP through setup wizard and agent selection to the shared workspace and a visible result.
03 / Benchmark
Prompt, assessment framework, run values, geodata, maps, charts and the portable QGIS project form one dossier.