Saved queries & dashboards
Because the model emits a plain query object, you can store it and re-run it later. That turns a one-off question into a dashboard widget that refreshes without the model in the loop. This page covers persisting a query, replaying it safely, and deriving its output shape.
A query is just data
The query the model produced is JSON. Persist it anywhere you’d store other application data:
await db.saveWidget(id, { query }) // it's just JSONReplay it without the model
To refresh the widget, run the stored query with valv.run. The model isn’t
involved, so there’s no token cost and no latency from a generation step:
const rows = await valv.run(widget.query, ctx)Every replay goes through the full pipeline: the stored query is re-validated and re-scoped to the current viewer’s context. A saved query is never trusted just because it ran once. It can’t outlive the permissions it was created under, and it returns each viewer only the rows their policy allows.
Derive the output shape
valv.resultSchema returns a query’s output columns and their types without
running it. Use it to configure a chart or to detect drift when your schema
changes:
const columns = valv.resultSchema(widget.query) // e.g. [{ name, type }, ...]