AI DASHBOARDS

Ask. The agent builds the dashboard.

The era of drag-and-drop self-serve BI is over. Sprinkle's agent knows your domain model, picks the right metrics and assembles a governed dashboard — KPIs, charts, drill-downs and all — that you can ship as-is or refine in plain English.

// how the agent works

It knows your model. It picks the visuals.

The agent reads your semantic layer, the tables behind it and the metrics you've already defined — then chooses the right cut, the right chart and the right drill path. You review, edit in plain English, and ship.

01 · understands

Grounded in your domain model.

The agent uses your warehouse schema, joins and the metric definitions you've already governed — no hallucinated columns, no invented numbers.

02 · assembles

Right metric, right chart, right cut.

Trend questions get time-series, comparison questions get bars, share-of-total gets a donut. The agent picks; you can override anything.

03 · refines

Edit by chatting, not by clicking.

"Split that by plan", "swap region for channel", "make it weekly." The agent updates the dashboard in place — every change is versioned and reviewable.

// grounded

An agent that actually knows your business.

Sprinkle's agent doesn't guess. It reads the semantic layer your team has already governed — entities, relationships, metric definitions — and uses that as its source of truth. Same numbers as the Finance dashboard. Every time.

  • Reads your warehouse schema and semantic layer at query time
  • Reuses governed metrics — never re-derives MRR
  • Respects row-level security and column masking
  • Shows its working: every chart links to the SQL it ran
// the build

Watch the agent assemble it.

From prompt to shipped dashboard takes seconds, and every step is visible. The agent narrates its choices — which metric it pulled, which chart it picked, which filter it applied — so you can intervene at any point.

  • Step-by-step reasoning, not a black box
  • Pin any step, replay it with a different choice
  • Generated SQL surfaced for every widget
  • Hand off the dashboard to a colleague mid-build
// refine

Iterate the way you think.

The first draft is rarely the last. Talk to the dashboard the way you'd talk to an analyst — split this, swap that, zoom in here. The agent applies the change, keeps the rest in place, and stays grounded in the same semantic layer.

  • Natural-language follow-ups update the live dashboard
  • Every change is versioned — rollback in one click
  • Suggested next questions, based on what you're looking at
  • Share a refinement thread with a teammate, not a screenshot
<5s
// prompt → assembled dashboard
0
// hallucinated metrics
100%
// grounded in your semantic layer
// follow-up refinements