Skip to main content

​
Voice Sessions

A voice session represents a single call between a user and an agent. Sessions track the full lifecycle from connection to post-call analysis.

​
Session Lifecycle

CREATED ---> ACTIVE ---> COMPLETED (triggers call analysis)
                    +--> FAILED
  1. CREATED: Session record created when a token is generated
  2. ACTIVE: User joins the LiveKit room and audio flows
  3. COMPLETED: Session ends normally, triggers call analysis
  4. FAILED: Session ends due to an error

​
Starting a Session

​
Browser (Preview/Test)

  1. User clicks “Preview” on an agent in the dashboard
  2. Frontend requests a LiveKit token from /api/livekit/token
  3. The token route creates the room, dispatches the agent, and creates a session record
  4. User connects via WebRTC and the agent joins

​
Phone (Inbound)

  1. Caller dials a Twilio number assigned to an agent
  2. Twilio routes the call through the SIP trunk to LiveKit
  3. LiveKit dispatch rule triggers the agent worker
  4. Agent fetches config from the backend and starts the pipeline

​
Phone (Outbound)

  1. User initiates a call from the dashboard via POST /api/calls/outbound
  2. Backend creates a LiveKit SIP participant for the phone number
  3. Agent joins the room and connects to the caller

​
Transcripts

Every message in the conversation is saved as a transcript entry with:
  • speaker: USER or AGENT
  • content: The message text
  • timestamp: When the message was spoken
Transcripts are streamed from the agent worker to the backend in real-time via POST /api/sessions/by-room/{room_name}/transcripts.

​
Post-Call Analysis

When a session ends, the backend triggers AI analysis using the full transcript. See Call Intelligence for details.

​
Viewing Sessions

Navigate to Dashboard → Call Logs to browse all sessions. Click any session to see:
  • Session metadata (status, duration, start time)
  • Full conversation transcript
  • Call analysis (summary, sentiment, topics, action items)
  • Cost breakdown by provider