01 / System
Plugin, bridge and four skills.
Versions of the three building blocks, layout of the working directory, the instruction file for the three CLIs and the four skills with before-and-after images.
Technical documentation
Configuration and reference for SpatialAgents.
This entry point covers agent profiles, local connection, shared workspace, artefacts and troubleshooting. The quickstart separately covers installation and first use; the plugin page explains the system architecture.
Installation and setup wizardSix steps from the plugin ZIP to the first visible QGIS task.
Agent profilesProject configuration for Codex CLI, Claude Code and OpenCode.
Local connectionBridge, token authentication, operations and approvals.
Workspace and artefactsShared files, project state, screenshots and outputs.
TroubleshootingSystematically check connection, CLI, project paths and result.
Reference
Three pages for the detail.
The reference describes the system, its QGIS side and the library used in the benchmark test setup.
Open system
02 / QGIS plugin
Connection, nine operation groups, approvals.
Prerequisites, port and token, the 67 tools by group, path rules, checksums, asynchronous jobs and file transfer.
03 / Geospatial API
The institute's Python library.
Research of the institute, not part of the SpatialAgents: objects, accessor pattern, nine data sources, thirteen algorithm categories and error messages that name the fix.
Agent profiles
Three CLIs, one shared domain context.
SpatialAgents support Linux; use on Windows is experimental. The selected CLI must be installed and authenticated first. The setup wizard then adds the project configuration, shared skills and access to the open QGIS project.
Codex CLI
Project-local configuration connects Codex to the SpatialAgents bridge and shared skills.
Official installation guide ↗Claude Code
Project instructions, skills and bridge configuration are available to Claude Code in the shared workspace.
Official installation guide ↗OpenCode
The project profile provides skills and bridge access for the selected local or container-based execution environment.
Official installation guide ↗Local connection
Bridge, authentication and QGIS operations.
The selected agent system starts the SpatialAgents bridge as a local process and makes it available to the participating agents. The bridge connects to the QGIS plugin with token authentication and transmits defined tool calls.
- Connection
- The local WebSocket connection joins the bridge to the running QGIS session.
- Authentication
- A bearer token protects communication between the two components.
- QGIS actions
- The plugin performs declared operations for project state, layers, navigation, styling and artefacts.
- Approvals
- Mutating plugin actions are presented for confirmation in QGIS.
Workspace and artefacts
Code, geodata and QGIS project in one place.
The agents and QGIS access the same project directory. Generated data, scripts, screenshots and project files therefore remain directly verifiable and reusable.
- Project configuration
- Agent profiles, skills and bridge settings reside in the project workspace.
- Geodata
- Rasters, vectors and tables are created as regular files and loaded into QGIS.
- Return path
- Layer information, project state and screenshots are available to the agents for subsequent checks.
- Artefacts
- The QGIS project, maps, charts and reports remain as traceable results.
Troubleshooting
From connection check to visible map.
For an unexpected result, first check the active QGIS session, then bridge and token, followed by the selected system configuration and workspace, and finally layers, map extent and generated files.
- Check the QGIS session.
Open the project, enable the plugin and confirm that the expected session is running.
- Check the connection.
Review the bridge process, token and connection reported by the plugin together.
- Check the workspace.
Start the selected CLI in the intended project directory and verify the profile and available skills.
- Check the result.
Read the layer tree, map extent, file paths and screenshot again before refining the task.