GeoMind

GeoMind is an AI-powered command-line agent for Sentinel-2 satellite imagery analysis. Type a plain English query - GeoMind geocodes the location, searches the satellite catalog, streams only the data chunks it needs, and produces images and statistics in seconds.

---

What GeoMind Can Do

Capability	Example Query
RGB true-color composite	"Create an RGB composite for the most recent image of Iceland"
NDVI vegetation index	"Show me NDVI data for the Amazon rainforest"
Cloud-filtered search	"Search for images with less than 10% cloud cover in London"
Band statistics	"Get band statistics for this scene"
Multi-product in one query	"Get me recent image of Scotland and its NDVI"
Coordinate-based query	"Get satellite data for coordinates 40.7128, -74.0060"
GeoCroissant metadata	"Get me any Iceland recent image geocroissant metadata"

A single query can trigger multiple tools automatically - search, composite, and NDVI all at once.

---

How It Works

Your query  ->  Geocoding  ->  Catalog search  ->  Stream band data  ->  Image output
"Paris RGB"     Paris coords    Recent scenes       ~1-5 MB via Zarr     outputs/*.png

1. Geocoding - place name converted to coordinates and a bounding box
2. Catalog search - recent Sentinel-2 scenes retrieved for that area
3. Cloud-native streaming - only the required band chunks are downloaded (~1–5 MB instead of ~720 MB)
4. Processing - bands are scaled, stacked, and rendered as PNG
5. Output - image saved to outputs/ and opened automatically

---

Installation

Available on PyPI

GeoMind is available on PyPI. A full release is coming soon.

pip install geomind-ai

Dependencies installed automatically: OpenAI client, PySTAC-Client, Zarr, Xarray, Geopy, Matplotlib, and more.

---

Quick Start

1. Launch GeoMind:

geomind

2. On first run, enter your free OpenRouter API key when prompted:

OpenRouter API key required (FREE)
   Get yours at: https://openrouter.ai/settings/keys

   Enter your API key: sk-or-v1-xxxxxxxxxxxxxxxxxxxxxxxx
   API key saved! You won't need to enter it again.

3. Start querying:

> get me recent image of scotland and its ndvi

Executing: list_recent_imagery({'location_name': 'Scotland'})
Executing: create_rgb_composite({...})
Executing: calculate_ndvi({...})

RGB composite:  outputs/rgb_composite_7066.png
NDVI:           outputs/ndvi_9664.png
NDVI stats:     min -724775 / max 161061 / mean -16.43 / std 2447.43

---

Key Facts

Data source	Sentinel-2 Level-2A (surface reflectance)
Resolution	10 m (RGB, NDVI), 20 m (Red Edge, SWIR)
Default AI model	nvidia/nemotron-3-nano-30b-a3b:free - free via OpenRouter
API key required	Yes - free OpenRouter account at openrouter.ai
Output format	PNG images saved to outputs/; GeoCroissant JSON-LD saved to outputs/
Metadata standard	GeoCroissant 1.0 + Croissant 1.1 - validated by mlcroissant validate
Data download per query	~1–5 MB (cloud-native streaming, no full scene download)
Search lookback	14 days by default, auto-extends if no results found
Max cloud cover	50% by default, configurable per query

---

Documentation

• Usage Examples

  ---

  Step-by-step walkthroughs - RGB composites, NDVI, cloud filtering, scripting, and more.

  Usage Examples

• CLI Reference

  ---

  All command-line flags, environment variables, and interactive session commands.

  CLI Reference

• Tools Reference

  ---

  Every tool the GeoMind agent can call - geocoding, STAC search, and image processing.

  Tools Reference

• Architecture

  ---

  The agent loop, cloud-native data flow, technology stack, and external APIs explained.

  Architecture

• GeoCroissant

  ---

  ML-ready metadata for every Sentinel-2 scene - vocabulary, output structure, validation, and examples.

  GeoCroissant