History & Logs

Monitoring conversation logs, recordings, transcripts, and debug traces.

View as Markdown

The Conversations Tab acts as the historical ledger for your business. Every Inbound, Outbound, and Web session is recorded here, providing a complete audit trail from the moment a connection is attempted to the moment it closes.

The Conversation List

This is your main view for monitoring traffic.

Status Indicators

Calls go through several states. Understanding them helps in troubleshooting.

StatusDescription
Queued(Outbound Bulk) The call is waiting for an available concurrency slot.
RingingConnection attempt in progress.
ActiveLive conversation happening right now.
CompletedSuccessfully finished.
FailedSystem error or Provider error (e.g., Invalid Number).
DeclinedUser rejected the call.
MissedUser did not answer within the timeout.

Filters

You can drill down into the data using filters:

  • Date Range: "Last 24 Hours", "This Month".
  • Type: Inbound vs Outbound vs Web.
  • Campaign: Filter by specific campaigns to analyze performance.

Conversation Detail View

Clicking on any row opens the Detail View. This view is split into three specific tabs.

1. Overview (Metadata)

The high-level summary of the session.

  • Session ID: The unique UUID (crucial for API debugging).
  • Cost: The calculated cost for this specific call.
  • Duration: Exact length in seconds.
  • Post-Analysis: Displays the generated Summary, Tags, and Extracted JSON Data.

2. Media & Transcript

The artifacts of the conversation.

  • Audio Player: Listen to the full MP3 recording. You can download this for quality assurance.
  • Transcript: A timestamped, speaker-diarized text log of who said what.
    • User: "Hello?"
    • Agent: "Hi, this is Sarah."

3. Debug Logs (Traces)

For Developers. This is the X-Ray view of the agent's brain. It shows the millisecond-by-millisecond execution flow.

If an agent felt "slow" or gave a "wrong answer," look here.

What you can see:

  • Latency Traces:
    • STT Final (User finished speaking) > LLM Request
    • LLM First Token (Time to think) > TTS Audio Stream
  • Tool Execution:
    • Inputs passed to the tool.
    • Raw JSON response received from the API.
    • Any errors generated by the JavaScript post-processor.
  • Interruption Events: Logs when a user barged in and how the VAD/AI handled it.

Debugging Latency

If you see a high duration between LLM Request and LLM First Token, check if you are using a slow model (e.g., GPT-4) or a complex system prompt. Switching to a faster model (e.g., GPT-4o-mini or Groq) usually fixes this.

On this page