> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aixy-gateway.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Monitors and notifications

> Turn usage signals into incidents and notify the people responsible.

Monitors evaluate traffic or budget signals and create incidents when a threshold is reached.
Use them to detect rising errors, slower requests, or an allowance approaching its limit.
Notification channels deliver incident and recovery messages to your team.

## Choose a signal

| Metric | Configuration |
| - | - |
| Budget utilization | Percentage of an existing budget used in its reset period |
| Error rate | Percentage of requests with an HTTP error |
| Error count | Number of failed requests in the observation window |
| P95 latency | End-to-end request latency in milliseconds at the 95th percentile |
| Success rate | Percentage of requests without an HTTP error |

Traffic monitors can focus on an organization, project, team, user, API key, provider, or model
within your permitted scope. A budget-utilization monitor selects an existing budget.

## Configure a monitor

1. Open **Monitors → Create monitor**, enter a name, and describe who owns the response.
2. Choose the metric, scope, target, comparison, and threshold.
3. Set the observation window and minimum request count to avoid reacting to an unrepresentative
   sample.
4. Set the evaluation interval, repeat-notification delay, and severity.
5. Select notification channels, enable the monitor, and save.

For example, a project error-rate monitor might trigger at 10% over 15 minutes with at least 20
requests. Choose thresholds that fit the application's normal traffic; these numbers are a setup
example, not a service guarantee.

## Add notification channels

Open the notification channels tab in **Monitors** and create an email, Slack, Discord, or webhook
channel. Enter the destination, save it, and use the channel test action to verify delivery before
relying on it. Tests send a real notification to the configured destination.

Select the saved channel in the monitor. An incident is recorded even when no delivery channel is
selected. Keep the repeat-notification delay long enough for the owner to respond.

## Investigate an incident

Review the monitor's state and last evaluation. **Insufficient data** means the sample requirement
was not met; an evaluation error is distinct from a healthy signal.

Open **Incidents**, inspect the measurement, and acknowledge the incident when someone takes
ownership. **Inspect activity** opens the related time window and available scope filters for
accounts with Activity access. Use the request evidence to decide whether to adjust a provider,
route, budget, or application. Recovery is evaluated separately from acknowledgement.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.