Monitoring and alerts

Set a threshold on any dimension and get told when quality drops.

What you'll have

Notification when a dimension crosses a line you set, instead of a dashboard nobody is watching.

You'll need

A project with evaluations arriving
A threshold you would actually act on

Nobody watches a dashboard. Set a threshold and get told instead.

Alerts

Set a threshold on any dimension or category. When scores cross it, you get notified.

The dimension-level threshold is usually the one you want. Overall score is an aggregate and moves slowly; grounding falling off a cliff while overall drifts down two points is the alert worth having.

Sessions

For multi-turn applications, sessions group turns from one conversation so coherence is measured across the exchange rather than per-message.