Widgets
A widget is a single chart or table on a custom dashboard. Each widget queries your platform data — traces, observations, costs, and scores — and renders it the way you choose.
ℹ
Widgets are created and configured within a dashboard. A widget you build can be reused across dashboards.
Chart types
| Type | Good for |
|---|---|
| Time series | A metric over time (cost, volume, latency, tokens) |
| Bar / bar list | Comparing a metric across a dimension (top models, users, agents) |
| Histogram | Distribution of a numeric value (e.g. latency or a numeric score) |
| Score aggregate | Aggregated quality scores (mean, distribution) |
| Table | Tabular breakdowns, including latency tables (p50/p95/p99) |
Configuring a widget
When you add a widget you choose:
- Data source / metric — what to measure (e.g. count, total cost, latency, a specific score).
- Dimensions — how to break it down (model, user, tag, environment, time bucket).
- Chart type — how to render it.
Dashboard-level filters apply on top, so the same widget reflects whatever slice the dashboard is filtered to.
Next steps
- Custom Dashboards — assemble widgets into a view.
- Scores — the quality data behind score widgets.