Srot
Verified Indian government geodata inside QGIS: 342 ISRO Bhuvan layers across 44 states, Census and Survey of India boundaries, and live data.gov.in feeds. Browse and add without any model; each layer is probed before loading and carries its own citation.
Srot turns plain-language requests into real QGIS work, with Indian geodata built in.
Ask for "the boundary of Pune district and every hospital in it" or "load LULC for Andhra Pradesh
and clip it to Guntur", and the agent resolves the place, fetches the data from verified Indian
sources, and runs the geoprocessing.
.
What makes it different from other AI plugins:
* India-first data catalogue - 342 ISRO Bhuvan WMS layers covering 44 states and union territories
(land use, village boundaries, slope, road and rail networks, hydrology, groundwater levels and all
23 river basins), Bhuvan satellite imagery, data.gov.in (live CPCB air quality, IMD gridded rainfall,
hospital directory, crop production, mandi prices), Census 2011 boundaries, and OpenStreetMap.
* A gazetteer that handles Indian naming in practice - Bengaluru/Bangalore, Prayagraj/Allahabad,
Varanasi/Banaras all resolve offline, and boundary files of different vintages are detected rather
than assumed, so asking for Telangana works on both the current and the pre-2014 district sets.
* Safety by construction - the agent never runs generated Python. Every action is a declared tool, and
Processing calls are validated against the algorithm registry before they execute.
* Zero third-party Python dependencies. Nothing to pip install. Uses only PyQGIS and Qt, so it installs
cleanly on Windows, macOS and Linux and honours your QGIS proxy settings.
* Works fully offline with a local Ollama model, or with Anthropic Claude, OpenAI, or any
OpenAI-compatible endpoint. API keys are stored encrypted in the QGIS Authentication Manager.
* Works with no model at all - the Browse tab lists the whole catalogue and adds any layer with one click,
needing no model and no provider. Only the data.gov.in feeds ask for a free portal key.
The natural-language tab is an optional layer on top.
* Reproducible - every session is recorded as a runnable PyQGIS script you can save, review and re-run.
* Citation-ready - each layer carries its publisher, licence, citation and retrieval date in the layer's
own metadata, and the whole project exports as a source list in plain text, Markdown or BibTeX.