Web Analytics Made Easy - Statcounter
Skip to content

Custom Dashboards

Custom dashboards give every team a live view of the signals they need to operate AI systems: traces, evals, cost, latency, model health, tool behavior, workflow status, and business KPIs.

What Dashboards Do

Dashboards combine widgets such as time series, tables, scorecards, histograms, heatmaps, trace lists, eval scorecards, cost views, and process-unit status panels.

When To Use Them

Use dashboards when you need a repeatable view for:

  • Agent reliability.
  • Model latency and quality.
  • Token spend and cost attribution.
  • Eval score trends.
  • Workflow health and SLA status.
  • Customer or product outcome monitoring.

Prerequisites

  1. Send data through a concrete integration path: OpenTelemetry, REST/HTTP events, JavaScript or pixel tracking, Claude Code, OpenAI Agents, or another ingestion method.
  2. Include stable dimensions such as workflow, agent, model, team, customer_tier, or business_unit.
  3. Decide who the dashboard is for: engineering, product, support, executives, or ML teams.

Step-By-Step Setup

  1. Open Dashboards.
  2. Create a dashboard or duplicate an existing template.
  3. Add widgets for the core question you want to answer.
  4. Filter by workflow, model, team, or environment.
  5. Add drill-down links to traces, evals, alerts, or process units.
  6. Save and share with the owning team.

Example Widget Definition

1
2
3
4
5
6
7
8
widget:
  type: time_series
  title: Token spend by workflow
  metric: cost_usd
  group_by: workflow
  filter:
    environment: production
  window: 24h

What Appears In The Console

Dashboards show live widgets with drill-down paths into the trace, model, eval, alert, pipeline, and process-unit records behind each metric.

Common Mistakes

  • Building one large dashboard for every persona.
  • Showing infrastructure health without AI quality or business impact.
  • Missing drill-downs from KPI to trace.
  • Grouping by fields that are unstable or missing.