Table of Contents
Alert Feature Overview
AITracer's alert feature allows you to instantly detect anomalies in your LLM applications and receive notifications.
Error Detection
Get notified when error rate exceeds threshold
Cost Monitoring
Get warned before budget overruns
Latency Monitoring
Detect response time anomalies
Usage Monitoring
Detect request count anomalies
Types of Alerts
1. Error Rate Alert
Triggers when the error rate exceeds the threshold within a specified time window.
| Setting | Description | Example |
|---|---|---|
| Threshold | Error rate percentage | 5% |
| Time Window | Evaluation period | 5min, 15min, 1hr |
| Minimum Requests | Minimum requests needed for evaluation | 10 requests |
2. Cost Alert
Triggers when daily/weekly/monthly costs exceed the threshold or are projected to exceed.
| Setting | Description | Example |
|---|---|---|
| Budget | Amount (USD) | $100 |
| Period | Daily/Weekly/Monthly | Monthly |
| Warning Threshold | Percentage of budget for warning | 80% |
3. Latency Alert
Triggers when P50/P95/P99 latency exceeds the threshold.
| Setting | Description | Example |
|---|---|---|
| Percentile | P50, P95, P99 | P95 |
| Threshold | Milliseconds | 3000ms |
| Time Window | Evaluation period | 15min |
4. Request Count Alert
Triggers when request count spikes or drops suddenly.
| Setting | Description | Example |
|---|---|---|
| Condition | Above/Below | Above |
| Threshold | Request count | 1000/hour |
| Time Window | Evaluation period | 1 hour |
Creating Alert Rules
Create from Dashboard
- Log in to the Dashboard
- Open the "Alerts" menu
- Click "+ New Rule" button
- Select alert type
- Configure conditions and thresholds
- Select notification channels
- Click "Save"
Filter Conditions
You can set the following filters for alert rules:
- Project: Target specific projects only
- Model: Target specific models (gpt-4, claude-3, etc.) only
- Provider: Target specific providers like OpenAI, Anthropic only
- Metadata: Filter by custom metadata
Notification Channels
Email Notifications
Available on all plans. Sends notifications to team members' email addresses.
- Instant notification when alert triggers
- Can send to multiple email addresses
- Notification frequency control (minimum interval setting)
Slack Integration
Available on Pro plan and above. Sends real-time notifications to Slack channels.
- Go to "Settings" -> "Integrations" in the dashboard
- Click "Connect to Slack"
- Authenticate your Slack workspace
- Select the notification channel
Includes alert name, trigger conditions, current value, and affected project/model.
Webhook
Available on Starter and above. Sends POST requests to any endpoint.
// Webhook payload example
{
"alert_id": "alert_abc123",
"alert_name": "High Error Rate",
"type": "error_rate",
"severity": "critical",
"triggered_at": "2025-01-15T10:30:00Z",
"condition": {
"threshold": 5,
"current_value": 8.5,
"time_window": "5m"
},
"filters": {
"project": "my-chatbot",
"model": "gpt-4"
}
}
Configuration Examples
Production Error Monitoring
Configuration:
- Alert Type: Error Rate
- Threshold: 5%
- Time Window: 5 minutes
- Minimum Requests: 20
- Filter: Project = production
- Notification: Slack #alerts channel
Monthly Cost Budget Monitoring
Configuration:
- Alert Type: Cost
- Budget: $500/month
- Warning Threshold: 80% ($400)
- Critical Threshold: 100% ($500)
- Notification: Email + Slack
Latency Anomaly Detection
Configuration:
- Alert Type: Latency
- Percentile: P95
- Threshold: 5000ms
- Time Window: 15 minutes
- Filter: Model = gpt-4
- Notification: Webhook (PagerDuty)
Plan Limits
| Feature | Free | Starter | Pro | Enterprise |
|---|---|---|---|---|
| Alert Rules | 3 | 10 | 50 | Unlimited |
| Email Notifications | 1 address | 5 addresses | Unlimited | Unlimited |
| Slack Integration | - | - | 1 workspace | Unlimited |
| Webhooks | - | 2 endpoints | 10 endpoints | Unlimited |
| Evaluation Interval | 15 min | 5 min | 1 min | Real-time |
Alerts for the same rule fire at minimum 5-minute intervals. This prevents notification spam.
