Which way to run it
The same .sql dashboards run everywhere below; what changes is who reads them, whether the data is live, and who may see which rows.
In the browserDuckDB-Wasm: open files, author, explore. Nothing to install.CLI, HTML and CIRender HTML and SVG, lint every pull request.A static websiteGitHub Pages, S3, nginx — no server to run.Authoring on your dataThe builder wired to your database, on loopback.The locked serverRead-only snapshots, refresh, reports and alerts.Login and row policiesOIDC, JWT, API keys; rows and columns per reader.EmbeddingSigned links and a web component in your own app.Agents and MCPA dashboard’s questions answered over REST and MCP.
Compare them
Section titled “Compare them”| Way to run it | Needs | Live data | Login & row policies | Readers interact | Best for |
|---|---|---|---|---|---|
| In the browser | nothing (DuckDB-Wasm) | files you open | — | ✔ | trying it, authoring, one-off analyses |
| CLI, HTML and CI | the dashboard CLI |
at render time | — | HTML: ✔ / SVG: — | reports, e-mail, CI checks |
| A static website | any static host (GitHub Pages, S3, nginx) | at build time, or Parquet fetched in the browser | — (the host’s own) | ✔ | public or intranet sites without a server |
| Authoring on your data | the full toolkit, loopback | ✔ live, read-write | — (local only) | ✔ (+ SQL editor) | building dashboards against your database |
| The locked server | the full toolkit, a reverse proxy | ✔ snapshot, refreshed | ✔ | ✔ | sharing dashboards with others |
| Embedding | a locked server | ✔ | ✔ (signed links, keys) | ✔ | dashboards inside your own app |
| Agents and MCP | a locked server in agent mode | ✔ | ✔ | — (questions and answers) | AI assistants and tools asking a dashboard’s questions |
Two rules of thumb:
- No server — a site built with
dashboard --siteruns entirely in the reader’s browser; there is no data API to secure, but everything you ship is public to whoever can open the site. - A server — when readers must only see their own rows, or the data must stay on your machines, run the locked server with login and policies. Never expose the authoring server beyond loopback.
The operator checklist covers what to set before going live.