starlet 0.2.0

Spatial tiling, MVT generation, and tile serving for geospatial data

Starlet

Spatial tiling, MVT generation, and tile serving for geospatial data.

Setup

python -m venv .venv
source .venv/bin/activate
pip install -e .

CLI

All commands are available through the starlet CLI.

starlet --help

starlet tile — Partition a dataset

starlet tile --input data.parquet --outdir datasets/mydata --num-tiles 40

Flag	Default	Description
--input	(required)	Path to GeoParquet or GeoJSON file
--outdir	(required)	Output dataset directory
--num-tiles	40	Target number of spatial partitions
--partition-size	1gb	Target partition size (e.g. 512mb, 1gb)
--sort	zorder	Sort order: zorder, hilbert, columns, none
--sample-cap	10000	Reservoir sampling cap for centroids
--compression	zstd	Parquet compression codec

starlet mvt — Generate vector tiles

starlet mvt --dir datasets/mydata --zoom 7 --threshold 100000

Flag	Default	Description
--dir	(required)	Dataset directory with parquet_tiles/ and histograms/
--zoom	7	Maximum zoom level
--threshold	0	Minimum feature count per tile
--outdir	<dir>/mvt/	MVT output directory

starlet build — Full pipeline (tile + MVT)

starlet build --input data.parquet --outdir datasets/mydata

starlet serve — Launch the tile server

starlet serve --dir datasets --port 8765

Flag	Default	Description
--dir	(required)	Root directory containing dataset subdirectories
--host	0.0.0.0	Host to bind
--port	8765	Port to bind
--cache-size	256	In-memory tile cache size

starlet info — Inspect a dataset

starlet info --dir datasets/mydata

Make Targets

Convenience wrappers around the CLI:

make tiles INPUT=path/to/data.parquet
make mvt   INPUT=path/to/data.parquet
make build INPUT=path/to/data.parquet   # tiles + mvt
make server                              # starts on port 8765
make clean                               # removes datasets/*

API Endpoints

Once the server is running:

Method	Path	Description
GET	/	Interactive dataset selector
GET	/api/datasets	List all datasets
GET	/datasets.json	Search datasets by name
GET	/datasets/<dataset>.json	Dataset metadata
GET	/datasets/<dataset>.html	Dataset detail page
GET	/<dataset>/<z>/<x>/<y>.mvt	Mapbox Vector Tile
GET	/datasets/<dataset>/features.<fmt>	Download features (csv/geojson)
POST	/datasets/<dataset>/features.<fmt>	Download with geometry filter
GET	/datasets/<dataset>/features/sample.json	Sample attributes
GET	/datasets/<dataset>/features/sample.geojson	Sample record with geometry
GET	/api/datasets/<dataset>/stats	Attribute statistics
POST	/datasets/<dataset>/styles.json	LLM-generated styling suggestions

LLM Styling Suggestions

The POST /datasets/<dataset>/styles.json endpoint uses an LLM to generate map styling rules from dataset attribute statistics.

Provider Selection

Set the LLM_PROVIDER environment variable to choose a provider:

export LLM_PROVIDER=gemini   # default
export LLM_PROVIDER=ollama   # local Ollama

Falls back to Gemini if the variable is unset or invalid.

Gemini (default)

Requires a Google AI Studio API key:

export GEMINI_API_KEY=your-key-here
starlet serve --dir datasets

Ollama (local)

Requires a running Ollama instance on the default port (11434):

ollama serve                  # start Ollama
ollama pull llama3            # pull a model (once)

export LLM_PROVIDER=ollama
starlet serve --dir datasets

To use a different model:

export OLLAMA_MODEL=mistral

See starlet/_internal/server/llm/README.md for full LLM provider documentation.

Example

# Full pipeline
starlet build --input ../data/TIGER2018_COUNTY.parquet --outdir datasets/TIGER2018_COUNTY

# Or via Make
make build INPUT=../data/TIGER2018_COUNTY.parquet

# Start the server
make server

Then open http://localhost:8765 and select a dataset to visualize.

Prerequisites

• Python 3.10+
• make (optional, for convenience targets)