Timeline
Turn-by-turn transcript, synced audio, and span waterfall.
The Timeline tab (/sessions/:sessionId/timeline) is the primary debugger for what happened
in a session, in order.
Conversational turns
A turn is one complete contribution by a participant (caller, AI agent, or human representative). Timeline lists turns chronologically with transcript text when available.
For voice, turn boundaries usually follow end-of-utterance from the voice pipeline. For text and messaging, each discrete message is typically a turn.
Audio playback
When recording captured session audio:
- Use the audio player to scrub the call
- Select a turn to jump playback to that segment when timing anchors are present
- Collapse the player when you only need the transcript
If no recording is available, Timeline still shows turns and spans; the audio zone explains why playback is missing.
Span waterfall
The span waterfall is the stage timing chart under each turn. Expand a turn (or use the detail panel) to see child work such as:
- Speech-to-text (STT)
- Large language model (LLM) calls
- Tool or API calls
- Text-to-speech (TTS)
Open a span for attributes and input/output previews (subject to
Gen AI content capture and PII redaction
settings). Shareable span URLs use /sessions/:sessionId/spans/:spanId.
Use the latency threshold control (when present) to highlight slow stages for quality review.
Speakers
Timeline distinguishes participants when speaker separation (diarization) or agent identity is available:
- Caller / user
- AI agents (by agent identity)
- Human representatives after escalation
Until post-session speaker separation finishes, some human audio may appear under a generic user lane.