Vercel ships custom observability monitors and AI alert validation to reduce false pages

Custom monitors built from Observability queries

Vercel has shipped two related additions to its Observability product: user-defined monitors that alert on arbitrary query results, and an AI validation step that decides whether an alert should actually page someone. Both announcements appeared on the Vercel blog on October 26.

According to Vercel, custom monitors turn any Observability query into an alerting rule. Suggested use cases include AI Gateway costs, Core Web Vitals and other performance metrics, failure rates for Workflows and external APIs, and CPU time consumed by Vercel Sandbox.

A monitor watches a single metric — response time, say — or a ratio between two metrics, such as errors over total requests. Users choose between statistical anomaly detection and a fixed threshold, then set the detection window, filters, and grouping. Vercel's own example tracks HTTP 500 errors as a share of all requests.

There are three ways to build one: inspect a chart's underlying query inside Observability, start from a preset, or open Settings > Alerts at the team level. Notifications can be delivered by email, Slack, Vercel Inbox, mobile push, or team-level webhooks.

AI validation before paging

The second feature targets alert fatigue directly. When validation is switched on for a webhook destination, Vercel Agent — the company's AI assistant — investigates each alert from a Vercel monitor before it reaches the paging system. Only alerts the agent rates as Critical are forwarded; the rest never trigger a page.

Setup is per destination: open a monitor's notification settings, create or link a webhook pointing at the paging tool (incident.io is Vercel's example), and pick the Critical Only option.

Vercel says Agent examines logs and metrics to judge actual impact. It can, for instance, recognise health-check failures that never touch real customer traffic — and so do not warrant waking anyone — or confirm that persistent 5xx responses are hitting checkout requests, which do. Teams can attach explanatory context to each monitor to steer the agent's investigation.

Availability and pricing

Both features require Pro or Enterprise plans with Observability Plus. Custom monitors are metered at $0.0035 per gigabyte of data scanned, while built-in alerts stay included with Observability Plus. Agent work is capped too: each billing cycle includes 10 investigations, with additional ones billed separately. Both announcements come from Vercel's own changelog, so capabilities and pricing are as described by the vendor.

Why it matters

Noisy paging is one of the most persistent problems in operations. Alerts that fire on health-check noise or transient blips train teams to distrust their monitoring, and genuine incidents get missed as a result. Handing triage to an AI layer that weighs severity before a page goes out is a grounded use of agents in ops: the system is not writing code, it is making a keep-or-escalate call based on logs and metrics.

The custom monitors close a real gap as well. Observability could previously surface data, but alerting was limited to built-in rules. Query-driven monitors that cover cost signals such as AI Gateway spend alongside error rates and Core Web Vitals reflect how spend has become a first-class operational metric for teams shipping AI features.

There are caveats. Severity classification is only as reliable as the investigation behind it, and an incident wrongly judged non-critical is silently held back from paging — so teams may want to keep an unvalidated channel for their most business-critical monitors at first. The metered pricing, both per gigabyte scanned and per investigation beyond ten per cycle, also means high-volume users should model costs before a wide rollout.