Skip to content
Pollenix Docs

Workbench → BI

Workbench → BI

The fastest path from “asked a question” to “published a widget.”

The flow

  1. In the Workbench, ask a question.
  2. Pollen8 runs the query, returns the result + an inferred chart spec.
  3. Click Send to BI in the answer’s footer.
  4. Pick a dashboard (or create a new one).
  5. Pick the widget kind to materialize as — the picker pre-selects what fits the data shape (line if time-series, bar if category × metric, donut for small breakdowns, kpi for single scalars, table otherwise).
  6. The widget lands on the canvas with the connection_id + SQL + chart spec baked in.

The widget refreshes when the user opens the dashboard or hits Refresh — it re-runs the SQL against the live warehouse.

What gets persisted

The widget owns:

  • connection_id — the workbench connection at the time of send.
  • spec_json.sql — the SQL the planner generated (or the hand-edited version if you edited).
  • spec_json.kind — the chart kind.
  • spec_json.x_field / y_fields[] — column bindings inferred from the result.
  • layout_json — default position; auto-placed at the next free spot on the canvas.

It does NOT own the original NL question. The dashboard is the SQL

  • spec, not the prompt — re-asking the question might produce slightly different SQL on a different day.

Edit after sending

Open the widget → Edit SQL opens the same editor as the workbench, scoped to that widget. Save → the widget updates in place. Refresh → re-runs.

Multiple widgets from one question

Sometimes one workbench answer is worth multiple widgets — say a table of patients + a KPI of the count. Send to BI twice, picking a different widget kind each time. Both widgets share the same SQL (if both are checked against the same connection); each renders its own view.

Why “send” not “auto-track”

Pollen8 doesn’t auto-create widgets from every workbench question. Most workbench questions are one-shot exploration. The explicit Send to BI action is the commitment that this question is worth materializing for ongoing monitoring.