Changelog

Unreleased

1.0.0

  • Fixed (#62): on macOS and Linux, every tool that touches the project (listing layers, zooming, selecting, …) reported Thread <id> result not found in runner in the chat, even though it had worked. Tool results now always reach the assistant, on every platform and QGIS version.

  • Model export: each processing result in the chat ends with Open in Model Designer and Save as .model3 links. The run becomes a QGIS processing model, with your layers as inputs and its steps wired together.

  • New setting, After a processing run: open each run in the Model Designer, or ask to save it, automatically.

  • Token usage: the log ends every request with the tokens it used, plus a session total.

  • More reliable processing:

    • QGIS checks the parameters before running;

    • near-miss layer names, option labels and parameter-name case are corrected automatically;

    • a task that failed only on a parameter retries with the same algorithm;

    • results are passed between steps by layer id;

    • result layers get unique names.

  • Fewer LLM calls: processing requests skip one call per task, since routing is decided while the request is split into tasks.

  • Web links in replies open in the browser.

  • Fixed: changing the model name, API key, Ollama settings or maximum tokens now applies to your next message; before, it only took effect after the provider, mode or temperature changed. The first message no longer sets up the model twice.

  • For developers: a headless test suite and this documentation site.

  • Security scan: resolved all Bandit findings reported by the QGIS plugin repository. Errors are no longer silently swallowed; where a failure is expected (a corrupted saved setting, an unwritable error-log file, a deleted Logs tab after reloading the plugin), it’s handled and logged.

  • The layer list given to the model now describes geometry-less tables and non-vector, non-raster layers (mesh, point cloud, …) correctly.

  • Qt6 compatibility: all QGIS enum values are now written in their scoped form (for example Qgis.MessageLevel.Critical), as the QGIS plugin repository’s Qt6 check requires. This works on QGIS 3 and QGIS 4.

  • Fixed: QGIS information messages mirrored into the Logs tab were labelled CRITICAL; they now keep their real level.

0.5.0

  • Added support for Anthropic models.

  • Saved settings, and a dependency check and installation function.

0.4.0

  • Added support for QGIS 4.x.

  • Fixed a UI bug.

0.3.2

  • Fixed an import issue.

  • Added support for multi-step geospatial workflows.

  • Added a confirmation dialog when removing a layer.

  • Fixed a state mutation issue.

  • Extended the geoprocessing agent to multiple processing providers (GRASS, GDAL, PDAL, …).

0.2.0

  • Improved UI help messages and user notifications.

  • Enhanced error handling for project operations.

  • Fixed a QGIS crash caused by illegal threading during project load, save, create and delete operations.

0.1.1

  • Fixed dependency installation issues on some systems.

0.1

  • Initial release: Ollama, ChatGPT and Gemini support; dependency installation; layer reading and writing; listing layers and columns; attribute and geometry queries; project create, load and save; experimental processing agent.