Alert Types

Reference for the four alert types available in Metrion.

Metrion supports four alert types, each monitoring a different aspect of your AI usage. When you create a rule, you choose one type and configure a threshold. Metrion evaluates the rule after every proxied request and sends an email when your usage reaches 90% (warning) or 100% (alert) of that threshold.

Budget

Monitors the total cost of your AI requests over a defined period.

Field	Details
Threshold unit	usd, eur, or chf
Providers	All providers, or specific ones (Anthropic, OpenAI, Gemini, Mistral, Grok)
Period	Start of current month, rule creation date, or a custom date — through today

The budget is measured in the currency you select. If you choose EUR or CHF, Metrion converts your USD costs at a fixed rate.

When to use it: Set a budget alert to catch unexpected spending early. For example, alert me when I’ve spent more than $50 on OpenAI this month gives you time to scale back before the bill arrives.

Error Rate

Monitors the percentage (or absolute count) of requests that return a 4xx or 5xx HTTP status from the provider.

Field	Details
Threshold unit	percent (0–100) or count (absolute number of errors)
Providers	All providers, or specific ones
Period	Start and end date for the measurement window

Use percent to catch systemic issues — e.g. alert me when my error rate exceeds 5%. Use count if you have a hard tolerance for a specific number of failures regardless of total request volume.

When to use it: Error rate alerts are useful for production workloads where reliability matters. A sudden spike in 4xx or 5xx responses often indicates a misconfiguration, a provider outage, or a rate-limit problem.

Latency p95

Monitors the 95th-percentile response latency across your requests, measured in milliseconds.

Field	Details
Threshold unit	ms
Providers	All providers, or specific ones
Period	Start and end date for the measurement window

The p95 value means that 95% of your requests are faster than the reported figure. It is a better indicator of real-world performance than the average, because it captures the tail of slow requests without being skewed by occasional outliers.

When to use it: Set a latency p95 alert when your application has response-time requirements. For example, alert me when my p95 latency exceeds 3000ms tells you when a meaningful share of your users is experiencing slow responses.

Request Volume

Monitors the total number of requests made through the Metrion proxy over a defined period.

Field	Details
Threshold unit	count
Providers	All providers, or specific ones
Period	Start and end date for the measurement window

When to use it: Request volume alerts are useful for staying within quota limits. For example, alert me when I’ve made 8,000 requests this month helps Free plan users stay under the 10,000-request monthly limit before they hit the cap. They’re also useful for cost forecasting and detecting unexpected traffic spikes.

Test each alert rule after creating it. Click Test on the rule to send a real notification to your account email immediately. Verifying delivery before you rely on the rule in production ensures you won’t miss a critical threshold because of a misconfigured email or a rule pointing at the wrong providers.