Where the data comes from
A dashboard’s setup statements are plain DuckDB SQL, so it reads anything DuckDB can read. What changes between the ways to run it is where that SQL runs, and therefore which files and networks it can reach.
| Source | In the browser (DuckDB-Wasm) | CLI, authoring server, locked server |
|---|---|---|
| CSV, Parquet, JSON files | files you open (Explore data), or URLs | local paths, a data folder |
| Files over HTTPS / S3 / GCS | HTTPS URLs whose server allows CORS | yes (httpfs) |
| A DuckDB database | — (open its tables as Parquet or CSV) | serve --db, ATTACH … (READ_ONLY) |
| Postgres, MySQL, SQLite | — | DuckDB’s database extensions |
| MotherDuck | — | ATTACH 'md:…' with a token |
On a locked server the dashboards read a read-only snapshot of the sources you give it, refreshed on a schedule, never the live database per viewer. See The locked server.