Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.blackbox.dasha.ai/llms.txt

Use this file to discover all available pages before exploring further.

Analyze completed calls in detail. The Call Inspector shows transcripts, audio recordings, timeline events, LLM prompts/responses, tool executions, and latency breakdowns for every conversation. What you’ll learn: Accessing the inspector, understanding sections, and navigating to detailed analysis.
Screenshots may differ from current UI version.

Access the Call Inspector

  1. Go to Calls in the sidebar
  2. Select the Past Calls tab
  3. Find your call in the list
  4. Click the Inspect button
  5. Inspector opens in a new tab
The Call Inspector is only available for completed calls. Calls in Pending, Queued, or Running status don’t have inspector data until they complete.

Inspector sections

SectionContentsUse caseLearn more
OverviewCall metadata, status, durationQuick contextBelow
TranscriptFull conversation with timestampsConversation review
AudioRecording playbackVoice qualityCommon Issues
LLMPrompts, responses, tokensModel debuggingCommon Issues
ToolsFunction calls and responsesIntegration debuggingCommon Issues
TimelineChronological eventsLatency analysis

Call overview

The inspector header displays key call metadata at a glance.

Header information

FieldDescriptionAPI field
Call IDUnique identifier (UUID)callId
StatusCompleted, Failed, or CanceledcallStatus
AgentAgent ID and nameagentId
DurationTotal conversation lengthdurationSeconds
EndpointPhone number or “web”endpoint
Call TypeConnection typecallType

Call type values

TypeAPI valueDescription
Inbound phoneInboundAudioIncoming phone call
Outbound phoneOutboundAudioOutgoing phone call
Web voiceWebCallBrowser WebRTC call
Web chatChatText conversation

Timestamps

TimestampMeaning
CreatedWhen the call was scheduled
StartedWhen the conversation began
CompletedWhen the call ended

Custom data

Any additionalData passed during call creation appears as formatted JSON:
{
  "callId": "660e8400-e29b-41d4-a716-446655440001",
  "status": "Completed",
  "duration": "00:03:45",
  "additionalData": {
    "customerId": "cust_789",
    "orderNumber": "ORD-12345"
  }
}

Quick analysis

Check conversation quality

  1. Open the Transcript section
  2. Scan for red flags: repetition, off-topic responses, user frustration
  3. Listen to audio for voice quality issues
  4. Review LLM logs if responses seem wrong
Common Issues

Check performance

  1. Open the Timeline section
  2. Look at latency breakdown: STT, LLM, TTS
  3. Identify the slowest component
  4. Compare against targets (under 2s total)
Common Issues

Debug failed calls

  1. Check Status in the header
  2. Look for error messages in system events
  3. Review LLM logs for API errors
  4. Check tool executions for failures
Common Issues

Best practices

During testing

  • Review 5-10 test calls before enabling an agent for production
  • Verify all conversation paths work as expected
  • Check that tool executions succeed
  • Confirm latencies are within acceptable ranges

In production

  • Sample 5-10% of calls randomly for quality review
  • Review all failed calls within 24 hours
  • Track performance trends over time
  • Monitor tool success rates

Data privacy

  • Redact PII before sharing transcripts externally
  • Comply with call recording consent laws in your jurisdiction
  • Implement data retention policies
  • Secure access to recordings with proper authentication

What’s next

Call History

View and filter past calls

Troubleshooting

Common issues and solutions

Common Issues

Solutions for frequent problems

Concurrency Monitoring

Monitor active calls and capacity