opengeos cli-anything-geoai
CLI harness for GeoAI -- AI-powered geospatial analysis from the command line.
Prerequisites
- GeoAI must be installed:
pip install geoai-py - PyTorch for model inference:
pip install torch torchvision - GDAL/rasterio for raster operations (installed with geoai)
Installation
cd agent-harness && pip install -e .
Verify: cli-anything-geoai --version
Global Options
| Option | Description |
|---|---|
--json |
Output as JSON (for agent consumption) |
--project PATH |
Load a project file for stateful operations |
--device cpu|cuda|mps|auto |
Force compute device |
--version |
Show version |
Command Groups
project -- Workspace management
cli-anything-geoai project new -n "analysis" -o project.json
cli-anything-geoai --project project.json project add-file image.tif
cli-anything-geoai --project project.json project info
cli-anything-geoai --project project.json project save
cli-anything-geoai --project project.json project list-files
cli-anything-geoai --project project.json project list-results
raster -- Raster inspection and operations
cli-anything-geoai raster info image.tif
cli-anything-geoai raster stats image.tif --band 1
cli-anything-geoai raster tile image.tif -o tiles/ -s 512 --overlap 64
cli-anything-geoai raster vectorize mask.tif -o polygons.geojson
vector -- Vector inspection and operations
cli-anything-geoai vector info buildings.geojson
cli-anything-geoai vector rasterize labels.geojson -t template.tif -o labels.tif -a class
data -- Data discovery and download
cli-anything-geoai data sources
cli-anything-geoai data search --bbox "-84.0,35.9,-83.9,36.0" -c sentinel-2-l2a
cli-anything-geoai data download naip --bbox "-84.0,35.9,-83.9,36.0" -o naip.tif
cli-anything-geoai data download overture --bbox "-84.0,35.9,-83.9,36.0" -o buildings.geojson
segment -- Image segmentation
# SAM automatic segmentation
cli-anything-geoai segment sam image.tif -o mask.tif
# Text-prompted segmentation (GroundedSAM)
cli-anything-geoai segment grounded-sam image.tif -o mask.tif -p "buildings"
# Semantic segmentation with a trained model
cli-anything-geoai segment semantic image.tif -m model.pth -o output.tif -n 5
# Train a segmentation model
cli-anything-geoai segment train -i images/ -l labels/ -o model_output/ --arch unet --backbone resnet50 --epochs 20
# List available models and architectures
cli-anything-geoai segment list-models
cli-anything-geoai segment list-architectures
detect -- Object detection
# Run detection
cli-anything-geoai detect run image.tif -m model.pth -n 5 -ov detections.geojson
# Train a detector
cli-anything-geoai detect train -i images/ -l labels/ -o model_output/ -n 5 --epochs 30
# List available models
cli-anything-geoai detect list-models
classify -- Image classification
# Train a classifier
cli-anything-geoai classify train -t train_dir/ -v val_dir/ -o model_output/ -m resnet50
# Predict
cli-anything-geoai classify predict image.jpg -m model.pth
change -- Change detection
# Detect changes between two images
cli-anything-geoai change detect before.tif after.tif -o changes.tif --method changestar
# List methods
cli-anything-geoai change list-methods
pipeline -- Batch processing
cli-anything-geoai pipeline run config.yaml -i input/ -o output/ -w 4
cli-anything-geoai pipeline show config.yaml
session -- State management
cli-anything-geoai session status
cli-anything-geoai session undo
cli-anything-geoai session redo
cli-anything-geoai session history
system-info -- Diagnostics
cli-anything-geoai system-info
Agent Usage Guide
JSON Output
Always use --json for machine-readable output:
result=$(cli-anything-geoai --json raster info image.tif)
JSON output returns either:
- Success:
{"key": "value", ...}(varies by command) - Error:
{"error": "message", "type": "ErrorType"}
Typical Agent Workflow
# 1. Inspect available data
cli-anything-geoai --json raster info satellite.tif
# 2. Check statistics
cli-anything-geoai --json raster stats satellite.tif --band 1
# 3. Run segmentation
cli-anything-geoai --json segment sam satellite.tif -o masks.tif
# 4. Vectorize results
cli-anything-geoai --json raster vectorize masks.tif -o buildings.geojson
# 5. Check results
cli-anything-geoai --json vector info buildings.geojson
Error Handling
All commands return clear error messages. In --json mode, errors are structured:
{"error": "Raster file not found: /path/to/file.tif", "type": "FileNotFoundError"}
Interactive REPL
Running cli-anything-geoai without arguments enters the REPL for interactive use:
> project new -n "my_analysis"
> raster info satellite.tif
> segment sam satellite.tif -o mask.tif
> session status
> quit