Monitor uptime for AI-native apps with Vercel Alerts

AI-native apps have unique reliability challenges. A single slow response from an upstream LLM API, a vector database timeout, or a spike in inference requests can cascade into 5xx errors across dozens of routes.

Vercel Anomaly Alerts acts a smoke detector for your app. The moment error rates or usage patterns deviate from your baseline, Vercel detects the anomaly, runs an AI investigation to identify the root cause, and notifies you with a summary of what's wrong before you have to dig through logs.

Vercel monitors your project continuously and detects two types of anomalies:

• 5xx error increase: Vercel detects a statistically significant spike in server errors on a route e.g. 75 failing requests in 5 minutes on /api/chat. Alerts resolve automatically once error rates return to baseline.
• Usage anomaly: Vercel detects an unexpected spike in invocations or data transfer relative to your recent traffic baseline.

When Vercel detects an anomaly, it notifies you immediately. When the AI investigation is enabled, Vercel also:

1. Groups similar alerts together, such as multiple routes affected by the same upstream issue.
2. Assigns a severity level based on the root cause and impact.
3. Generates a summary of the root cause and impact, so you know what's wrong at a glance.

Alerts are available on Pro and Enterprise plans with Observability enabled.

To set up alerts for your project, go to the Observability Alerts section of your project or follow these steps:

1. Open your project in the Vercel Dashboard.
2. Go to Observability → Alerts.
3. Select Enable alerts.

Vercel starts monitoring your project immediately. You'll receive email notifications by default.

To enable AI-powered alert grouping, severity assignment, and root-cause summaries, enable Agent investigations separately by following the steps in Getting started with Agent Investigation.

Observability Plus includes 10 investigations per billing cycle. Additional investigations are billed at $0.30 plus token costs.

For faster response times, route alert notifications directly to your team's Slack channel or your own incident management tooling.

1. In your project's Alerts settings, select Add notification.
2. Choose Slack and follow the OAuth flow to connect your workspace.
3. Select the channel where alerts should be posted.

1. Select Add notification → Webhook.
2. Paste your webhook URL.

Vercel sends a POST request to your webhook whenever an alert group is created, updated, or resolved.

Not every anomaly requires immediate action. By default, Vercel notifies you of all alerts. If you have AI investigation enabled, you can create an Alert Rule that filters to high-severity alerts only to reduce noise.

1. Go to Observability → Alerts → Rules.
2. Select Create rule.
3. Enable high alerts only.
4. Assign the rule to a specific notification channel (for example, a dedicated #incidents Slack channel).

Severity is determined by the AI investigation agent. High severity indicates significant impact, such as a large percentage of requests failing or sustained error rates. Lower-severity alerts remain visible in the Alerts dashboard.

When an alert fires, you receive a notification with:

• Alert type: Whether it's a 5xx error spike or a usage anomaly.
• Affected routes: The specific endpoints or pages where errors were detected.

If AI investigation is enabled, the notification also includes:

• AI-generated title: A plain-language summary of the root cause, e.g. Redis auth failure: 92% increase in errors on /api/chat.
• Severity: low, medium, or high.

You can click through to the Alerts dashboard to view the full investigation, including the agent's findings, related alerts in the same group, and a timeline of when errors started and stopped.

Vercel groups related alerts together to reduce noise. How grouping works depends on whether AI investigation is enabled:

• Without AI investigation: Alerts triggered within a 10-minute window are grouped together.
• With AI investigation: AI determines which alerts are related, including alerts triggered on different dates, and groups them into a single incident.

Either way, you get one grouped notification with the full picture instead of a separate notification for each affected route.

AI apps often have variable traffic patterns: usage can spike sharply during promotions, bot activity, or viral moments. Vercel uses a full week of historical data as the baseline for usage anomaly detection, which accounts for regular weekly patterns like higher weekday traffic.

If you're still seeing false positives, you can:

• Create rules that filter alerts by HTTP status code group such as only alert on 5xx.

• View your project's alert history in Observability → Alerts.
• Configure Alert Rules to control which alerts reach your team and how.
• Set up a webhook to connect Vercel Alerts to your incident management tool.