Skip to main content

What are traces?

Traces let you see the complete flow of a call broken down into individual steps called spans. Each span represents a specific operation — like a conversation turn, an agent response, or a tool execution — with its exact duration. While call logs tell you what happened, traces tell you how long each step took and how they relate to each other.

Trace structure

Each call generates a trace with a hierarchical structure. The root span represents the entire session, and individual steps are nested within it:
Agent session (full call)
├── User turn
│   └── User speaking
├── Agent turn
│   ├── LLM generation (AI response)
│   ├── Voice synthesis (text to audio)
│   ├── Tool executed (if applicable)
│   └── Agent speaking
├── Path transition (if using conversational paths)
└── ...more turns

Span types

SpanWhat it represents
Agent sessionThe entire call from start to finish
User turnA complete turn of the contact speaking
Agent turnA complete turn of the agent responding
LLM generationThe time the AI took to generate a response
Voice synthesisThe time to convert text to audio
Tool executedThe execution of a tool (integration, transfer, etc.)
Path transitionA node change in agents with conversational paths

Information available in each span

Depending on the span type, you can see different attributes:

Conversation turns

AttributeDescription
User messageWhat the contact said
Agent responseWhat the agent replied
InterruptedWhether the contact interrupted the agent

Executed tools

AttributeDescription
Tool nameWhich tool was executed
ArgumentsThe parameters it was called with

Path transitions

AttributeDescription
Source nodeWhich node the transition was made from
Destination nodeWhich node the agent moved to
Node typeThe type of the destination node

Use cases

Diagnose slow calls

Identify which step is taking the longest: AI generation? Voice synthesis? An external integration?

Verify conversational flows

Confirm that conversational paths follow the expected transitions between nodes.

Analyze tools

Review which tools were executed, with what parameters, and how long they took.

Understand interruptions

Detect when contacts interrupted the agent and how it responded.