Copy page

Core Concepts

Monitoring

Neuratel provides real-time monitoring so you can observe your voice agents in production. Track active calls, stream platform events, and watch quality metrics as they happen.

SSE Event Stream

The platform event stream delivers real-time events via Server-Sent Events (SSE). This is the primary channel for observing what is happening across your organization at any moment.

This endpoint is not in the API Reference tab. Connect to it directly as shown below.

Code
curl -N https://api.neuratel.ai/stream/platform \
  -H "Authorization: Bearer nk_live_your_api_key"

Code
import os, httpx

with httpx.stream(
    "GET",
    "https://api.neuratel.ai/stream/platform",
    headers={"Authorization": f"Bearer {os.environ['NEURATEL_API_KEY']}"},
) as response:
    for line in response.iter_lines():
        if line.startswith("data:"):
            print(line)

Code
const eventSource = new EventSource(
  "https://api.neuratel.ai/stream/platform",
  { headers: { Authorization: `Bearer ${process.env.NEURATEL_API_KEY}` } }
);

eventSource.onmessage = (event) => {
  const payload = JSON.parse(event.data);
  console.log(payload.event, payload.data);
};

Events Delivered

Event	Description
call.started	A new call has begun
call.ended	A call has finished — includes duration, outcome, and metrics
call.ringing	Outbound call is ringing at the destination
call.answered	Call has been picked up
call.failed	Call failed to connect
call.transferred	Call was transferred to a destination
agent.turn.started	Agent started processing a response
agent.turn.ended	Agent finished speaking
agent.tool.called	Agent invoked an external tool during the conversation
campaign.started	A campaign began dialing
campaign.completed	All contacts in a campaign have been processed
notification	In-app notification created

---

Active Calls

Get a snapshot of all calls currently in progress across your organization:

Code
curl https://api.neuratel.ai/v1/voice-sessions/active \
  -H "Authorization: Bearer nk_live_your_api_key"

Code
import os, httpx

response = httpx.get(
    "https://api.neuratel.ai/v1/voice-sessions/active",
    headers={"Authorization": f"Bearer {os.environ['NEURATEL_API_KEY']}"},
)
active = response.json()
print(f"{active['total_active']} calls in progress")

Code
const response = await fetch("https://api.neuratel.ai/v1/voice-sessions/active", {
  headers: { Authorization: `Bearer ${process.env.NEURATEL_API_KEY}` },
});
const active = await response.json();
console.log(`${active.total_active} calls in progress`);

Response:

Code
{
  "calls": [
    {
      "id": "550e8400-e29b-41d4-a716-446655440000",
      "channel": "phone",
      "direction": "inbound",
      "phone_number": "+14155559999",
      "agent_name": "Support Agent",
      "agent_id": "660e8400-e29b-41d4-a716-446655440001",
      "status": "in-progress",
      "started_at": "2025-01-15T14:30:00Z",
      "duration_seconds": 45
    }
  ],
  "total_active": 1,
  "timestamp": "2025-01-15T14:30:45Z"
}

---

WebSocket Metrics

For continuous metric updates (e.g., dashboards or monitoring consoles), connect to the WebSocket metrics endpoint:

Code
wss://api.neuratel.ai/ws/metrics

The WebSocket delivers periodic metric snapshots including:

Metric	Description
active_calls	Current number of in-progress calls
calls_per_minute	Calls initiated in the last rolling minute
avg_latency_ms	Average response latency across active calls
error_rate	Percentage of calls that ended in error over the last 5 minutes

Code
const ws = new WebSocket("wss://api.neuratel.ai/ws/metrics");

ws.onmessage = (event) => {
  const metrics = JSON.parse(event.data);
  updateDashboard(metrics);
};

Code
import asyncio, websockets, json, os

async def monitor():
    uri = "wss://api.neuratel.ai/ws/metrics"
    async with websockets.connect(uri) as ws:
        async for message in ws:
            metrics = json.loads(message)
            print(f"Active: {metrics['active_calls']}, Latency: {metrics['avg_latency_ms']}ms")

asyncio.run(monitor())

---

Quality Metrics

Monitor the quality of your voice agents across three dimensions.

Audio Quality

Audio quality is tracked at the media layer. Indicators include:

Metric	Healthy Range
Packet loss	< 1%
Jitter	< 30ms
Round-trip time	< 200ms

Audio quality issues typically point to network conditions between the caller and the media server, not your agent configuration.

Call Duration Distribution

Track whether calls are ending too quickly (possible misconfiguration or bad prompts) or running too long (possible conversation loops). Use the volume endpoint to spot outliers:

Code
GET https://api.neuratel.ai/v1/voice-sessions?duration_lt=10&status=completed

Calls under 10 seconds that completed successfully are worth investigating — the agent may be hanging up prematurely.

Error Rates

Monitor error rates to catch systemic issues early:

Code
GET https://api.neuratel.ai/v1/voice-sessions/analytics/summary?startDate=2025-01-15&endDate=2025-01-15&groupBy=agent

Check the failed_calls and success_rate fields per agent. A success rate below 80% warrants immediate investigation.

---

Health Check

A simple health endpoint for uptime monitors and load balancer checks:

Code
curl https://api.neuratel.ai/health

Returns 200 OK when the API is operational.

---

Dashboard Monitoring

The Neuratel dashboard provides real-time visibility without writing code:

• Active Calls — live list of all in-progress calls with agent names, durations, and phone numbers
• Call Volume — real-time chart of calls per hour
• Agent Activity — which agents are currently handling calls
• Recent Calls — latest completed calls with outcomes and metrics

---

Alerting

Set up alerts through two channels:

In-app notifications — the platform generates notifications for failed calls, agent errors, and campaign completions. See Notifications.

Status subscriptions — subscribe to email or Slack alerts for platform-wide incidents. See Status Page.

For custom alerting logic, consume the SSE event stream and trigger alerts in your own systems when conditions are met (e.g., error rate exceeds threshold, latency spikes above acceptable limits).

---

Next Steps

• Notifications — In-app event notifications
• Analytics — Historical metrics, trends, and post-call analysis
• Status Page — Platform availability and incident history
• Calls and Sessions — Per-call details and supervisor controls