Latency

The Latency tab (/sessions/:sessionId/latency) focuses on how long each stage of the conversation took.

What it covers

Depending on channel and how your agent is instrumented, you may see:

• End-to-end turn latency
• Large language model (LLM) time-to-first-token and generation time
• Speech-to-text (STT) and text-to-speech (TTS) delays
• Tool call wait time
• Dead air or gaps between turns

For highlighting slow stages on the turn list, use the latency threshold control on Timeline.

Goal and quality context

Use Quality for outcome, goal, sentiment, and evaluation results. Pair slow turns here with those signals to decide whether delay contributed to abandonment or escalation.

For fleet-level speed trends, start on Overview or the agent workspace Latency tab, then drill into individual sessions here.

Related settings

Accurate voice timing benefits from recording anchors. Pipeline content visibility depends on Gen AI content capture.