Call Monitoring

The Call Monitoring dashboard provides real-time and historical views of voice conversations. Review transcripts, analyze performance metrics, and debug agent behavior.

Accessing Call Monitoring

Navigate to Calls in the dashboard sidebar to view all calls.

Calls List View

The calls list displays all conversations with filtering and search:

Columns

Filtering

Filter calls by:

• Status - View only active, completed, or failed calls
• Workflow - Filter by specific workflow
• Date range - Set time boundaries
• Search - Find calls by ID or user

Status Indicators

Call Detail View

Click on any call to view its details.

Header Section

• Call ID - Unique identifier (click to copy)
• Status Badge - Current call status
  • Active calls show a pulsing indicator
• Refresh Button - For live calls, manually refresh data

Metrics Cards

Five performance metrics are displayed:

Duration

Total call length in minutes and seconds.

STT Latency

Speech-to-text processing time (average):

• Green (< 300ms) - Excellent
• Yellow (300-600ms) - Acceptable
• Red (> 600ms) - Needs attention

LLM Latency

Language model response time (average):

• Green (< 500ms) - Excellent
• Yellow (500-1000ms) - Acceptable
• Red (> 1000ms) - Needs attention

TTS Latency

Text-to-speech generation time (average):

• Green (< 200ms) - Excellent
• Yellow (200-500ms) - Acceptable
• Red (> 500ms) - Needs attention

Tokens Used

Total LLM tokens consumed during the call.

Info Cards

Three information cards provide context:

• Workflow - Link to the workflow used
• User - User identifier (if available)
• Timeline - Start and end timestamps

Agent Flow

Visualizes the sequence of agents that handled the call:

• Step numbers - Order of agent involvement
• Agent badges - Each agent in the flow
• Start/End labels - Entry and exit points
• Repeat indicators - Shows if an agent was called multiple times

The flow uses a connected timeline view:

• Desktop: Horizontal flow with connecting line
• Mobile: Vertical stack

Transcript Section

The full conversation transcript with two view modes:

Chat View

Interactive chat-style display with:

• User messages - Right-aligned, blue background
• Agent messages - Left-aligned, gray background
• Agent badges - Shows which agent responded
• Timestamps - Hover for full date/time

Raw Data View

JSON representation of the transcript for debugging.

Transcript Viewer Features

The enhanced transcript viewer provides powerful tools for analysis.

Search

Type in the search box to find specific content:

• Searches message text
• Searches agent names
• Matches are highlighted in yellow

Filtering

Click the Filter button to show/hide:

By Speaker

• User Messages - Toggle user utterances
• Agent Messages - Toggle agent responses

By Agent

When multiple agents participated, filter by specific agents:

• Check/uncheck individual agents
• See only relevant portions of the conversation

Export

Export transcripts in multiple formats:

Plain Text (.txt)

[10:30:45 AM] User: I'd like to log my breakfast
[10:30:48 AM] MealAgent: Sure! What did you have?

JSON (.json)

Full structured data including all metadata.

CSV (.csv)

Spreadsheet-compatible format with columns:

• Timestamp, Speaker, Agent, Message

Copy to Clipboard

Click the copy button to copy the filtered transcript to your clipboard.

Message Expansion

Long messages are truncated by default:

• Click Show more to expand
• Click Show less to collapse
• Threshold: 300 characters

Results Counter

Shows how many messages match your current filters:

12 of 45 messages

Live Call Monitoring

For active calls:

1. Status badge shows pulsing indicator
2. Refresh button appears in header
3. Click refresh to update transcript and metrics
4. Auto-scroll keeps newest messages visible

Analytics Dashboard

Navigate to Analytics for aggregate metrics:

Key Metrics

• Total calls over time
• Average call duration
• Completion rate
• Agent utilization

Performance Trends

• Latency trends by provider
• Token usage patterns
• Peak usage times

Debugging Workflows

Use call monitoring to debug workflow issues:

Check Agent Flow

1. Open call detail
2. Review agent flow visualization
3. Verify expected agent sequence
4. Look for unexpected repeats

Analyze Transcript

1. Filter by specific agent
2. Search for error keywords
3. Check tool execution results
4. Verify handoff messages

Review Metrics

1. Compare latency to baselines
2. Check for timeout issues
3. Identify slow providers
4. Monitor token consumption

Best Practices

Regular Monitoring

1. Check failed calls daily - Identify patterns early
2. Review long calls - May indicate confusion
3. Monitor latency trends - Catch degradation
4. Sample successful calls - Verify quality

Performance Optimization

1. Track baseline metrics - Know what's normal
2. Alert on thresholds - Set up monitoring
3. Compare across workflows - Find bottlenecks
4. Test after changes - Verify improvements

Quality Assurance

1. Review random samples - Check agent responses
2. Export for analysis - Detailed review offline
3. Track completion rates - Measure success
4. Gather user feedback - Correlate with transcripts