Your app is just three files — index.html, layers-input.json, and system-prompt.md. The core library loads from CDN.
STAC + SQL analytics
Point at a STAC catalog; the agent queries H3-indexed Parquet via DuckDB/MCP and controls the map in response.
Reports you can re-run
Save a session as a Quarto-style report — the SQL the agent ran as R, Python or SQL chunks with their output, charts, the final map, and a disclosure of the model used.
Flexible deployment
GitHub Pages with user-supplied API keys, Hugging Face Spaces with a secret config, or Kubernetes with server-injected credentials.