Monitor an agent's conversation clusters

Agent metric and evaluation monitors can now target a conversation cluster — either an intent cluster or an issue cluster.

Two ways to use it.

• Filter by cluster runs a targeted evaluation over only that cluster's conversations, or tracks how that group performs over time.
• Segment by cluster produces one monitor with a series per cluster, and new clusters arrive as new series on their own.

An evaluation judge could previously be pointed at a workflow but never at a cluster, so a refund request was scored the same way as a question about the return policy. It also allows alerting on cluster volume: refunds at 50% of traffic mean nothing unless you know they were 3% last month.

Find it in the agent metric or agent evaluation monitor creation flow for any agent with clustering enabled.