Metrics - Cartesia Docs

Metrics score completed calls in the Playground and through the API. Use built-in metrics for runtime health and custom metrics for outcomes specific to your application.

Built-in metrics

Cartesia calculates these metrics for every call:

Metric	Description
`system_call_success`	Whether the call ended without an agent failure.
`system_text_to_speech_ttfb`	Time to the first byte of generated audio on the first assistant turn.

Custom metrics

Create an LLM-as-a-judge metric in the Playground with a name and evaluation prompt. The prompt should define one concrete question and the expected output. Test it against representative call transcripts before assigning it to an agent.

Determine whether the caller's issue was resolved.
Return a boolean value and one sentence explaining the result.

Metric names must be unique within your organization. They may contain lowercase letters, numbers, hyphens, underscores, and periods.

Assign a metric

Assign metrics from the agent page in the Playground. Cartesia evaluates subsequent calls for that agent; assigning a metric does not evaluate earlier calls.

See Results to retrieve and export metric results.

​Built-in metrics

​Custom metrics

​Assign a metric

Built-in metrics

Custom metrics

Assign a metric