1. Check agent status → Must be "Active"
2. Verify API key → Not expired
3. Test in Chat mode → Isolate voice issues
4. Check knowledge base → Not empty

1. Go to Agents tab
2. Select affected agent
3. Toggle status to "Inactive"
4. Wait 5 seconds
5. Toggle back to "Active"
6. Test again

1. Open agent settings
2. Verify:
   - System prompt exists
   - Model is selected
   - Voice is configured
   - First message is set
3. Save and test

1. Go to API Keys
2. Check expiration date
3. Verify permissions include "agent:read"
4. Generate new key if needed
5. Update in your integration

Fix:
1. Check Numbers tab
2. Verify number is active
3. Ensure assigned to correct agent
4. Test with different number

Fix:
1. Test voice in Voice tab
2. Select different voice
3. Reset voice settings to defaults:
   - Stability: 0.75
   - Speed: 1.0
4. Contact support if persists

Fix:
1. Check "End call on silence" setting
2. Set to -1 (never) for testing
3. Gradually adjust to find optimal value

Current Models by Speed:
1. Gemini 2.5 Flash Lite (400-500ms) - Fastest
2. Gemini 2.5 Flash (500-700ms) - Fast
3. GPT-4o (1-2s) - Balanced
4. Claude 4 Sonnet (1-3s) - Advanced

1. Remove unnecessary content
2. Structure information clearly
3. Use categories effectively
4. Limit to essential information
5. Update indexing:
   - Delete and re-add content

Before:
"You are an assistant who helps customers with various
queries about our services, products, pricing, and can
also help them with booking appointments..."

After:
"You are a booking assistant for [Business].
Main tasks: 1) Answer service questions 2) Book appointments.
Be concise and direct."

Reduce timeout gradually:
- Start: 7 seconds
- Test: 5 seconds
- Optimal: 3-5 seconds for most cases

1. Check your internet speed (min 10 Mbps)
2. Use wired connection if possible
3. Close bandwidth-heavy applications
4. Test from different network

1. Select closest model region
2. Use CDN-enabled content
3. Consider edge deployment

1. Go to Conversations tab
2. Filter failed conversations
3. Identify patterns in misunderstandings
4. Note specific phrases causing issues

Check for:
- Missing information
- Outdated content
- Conflicting data
- Unclear formatting

Add specific instructions:

"IMPORTANT CONTEXT:
- Our business name is [Name]
- We specialize in [Services]
- Common questions include [List]

When unsure, ask for clarification rather than guessing."

Include in system prompt:

"Example conversations:
Customer: 'Do you have tables available?'
You: 'I'd be happy to check availability.
      For what date and time?'

Customer: 'What's your specialty?'
You: 'Our specialty is [specific dish/service].
      Would you like to know more about it?'"

Natural Settings:
- Stability: 0.65-0.75 (not too high)
- Similarity: 0.70-0.80
- Speed: 0.95-1.05
- Speaker Boost: Off for natural sound

Temperature Adjustments:
- Too rigid (0.0-0.2): Increase to 0.4-0.6
- Too random (0.8-1.0): Decrease to 0.5-0.7
- Optimal range: 0.4-0.6 for most cases

Add personality instructions:

"CONVERSATION STYLE:
- Use natural, conversational language
- Include appropriate fillers ('um', 'let me check')
- Vary your responses
- Show empathy and understanding
- Use the customer's name when known"

1. Increase turn timeout:
   - Current: 3 seconds → Try: 5-7 seconds
2. Change detection mode:
   - Switch from "Silence" to "Server VAD"
3. Test with various speakers

"IMPORTANT: Let customers finish speaking completely.
Wait for a natural pause before responding.
If unsure whether they're finished, wait an extra moment."

// Open browser console (F12)
// Look for errors like:
// - CORS errors
// - 404 not found
// - Authentication failures

<!-- Correct placement (before </body>) -->
<script>
  window.voisaConfig = {
    agentId: 'agt_YOUR_ID',
    position: 'bottom-right'
  };
</script>
<script src="https://cdn.voisa.ai/widget.js"></script>
</body>

1. Go to Agent Settings
2. Add your domain to "Allowed Hosts"
3. Include variations:
   - https://yourdomain.com
   - https://www.yourdomain.com
   - http://localhost:3000 (for testing)

curl -X GET https://api.voisa.ai/v1/api/verify \
  -H "Authorization: Bearer YOUR_API_KEY"

1. Expired key → Generate new one
2. Wrong environment → Check production vs staging
3. Missing "Bearer" prefix → Include in header
4. Incorrect permissions → Update key permissions

1. Go to MCP Tools
2. Click on server
3. Use "Test Connection"
4. Check response

1. Edit agent
2. Go to Tools section
3. Ensure tools are checked
4. Save agent

Review tool configuration:
{
  "permissions": {
    "execute": true,
    "max_executions": 10
  }
}

1. Test different voices
2. Use "Professional" tier voices
3. Enable Speaker Boost for clarity

Requirements:
- Bandwidth: 128 kbps minimum
- Latency: < 150ms
- Packet loss: < 1%

1. Clear voice cache:
   - Save agent without voice
   - Re-select voice
   - Save again
2. Verify voice ID in API response
3. Check voice provider status

1. Privacy Settings:
   - Zero Retention Mode: Must be OFF
   - Delete Transcripts: Must be OFF
2. Retention Days: > 0
3. Storage quota not exceeded

1. Enable Privacy Mode:
   - Set retention to 7-30 days
   - Enable auto-deletion
   - Disable audio storage

2. Data Export:
   - Use API endpoint: GET /api/export/user/{id}
   - Provide within 30 days

3. Right to Deletion:
   - API: DELETE /api/user/data/{id}
   - Confirm deletion in 72 hours

window.voisaConfig = {
  agentId: 'agt_YOUR_ID',
  debug: true,
  logLevel: 'verbose'
};

curl -X GET https://api.voisa.ai/v1/api/agents/list \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "X-Debug-Mode: true"

// Run in browser console
const diagnostics = {
  browser: navigator.userAgent,
  connection: navigator.connection,
  memory: performance.memory,
  timing: performance.timing,
  errors: window.__voisaErrors || []
};
console.log(JSON.stringify(diagnostics, null, 2));

// Add to system prompt for testing
"DEBUG: Log timestamp before and after processing.
Example: [START: 10:30:45.123] Processing query...
        [END: 10:30:46.789] Response ready."

1. Account ID: acc_xxxxx
2. Agent ID: agt_xxxxx
3. Timestamp of issue: 2025-01-15 10:30 UTC
4. Error messages (exact text)
5. Steps to reproduce
6. Screenshots/recordings
7. Browser console logs
8. Network traces (if applicable)

1. Go to Settings → Notifications
2. Configure alerts for:
   - Error rate > 10%
   - Response time > 3s
   - Failed calls > 5/hour
   - API errors

# Automated health check script
#!/bin/bash
curl -X GET https://api.voisa.ai/v1/api/verify \
  -H "Authorization: Bearer $API_KEY" \
  --fail --silent --show-error