---
type: "post"
id: "vercel-observability-adds-custom-monitors-with-threshold-and-bb5dca"
slug: "vercel-observability-adds-custom-monitors-with-threshold-and-bb5dca"
title: "Vercel Observability adds custom monitors with threshold and anomaly alerts"
description: "Vercel Observability now lets teams create custom monitors from query results, with fixed-threshold or statistical-anomaly detection."
retrieval_nugget: "Vercel Observability now lets teams create custom monitors from query results, with fixed-threshold or statistical-anomaly detection. The release matters to AI application teams because spend, sandbox consumption, workflow failures, and external API reliability can be treated as monitored operational signals rather than values someone checks manually in a dashboard."
published_at: "2026-10-27"
updated_at: "2026-10-27"
record_date: "2026-10-27"
date_kind: "published_at"
topics: ["agent-security","sandbox"]
entities: ["Vercel","Vercel Observability","Vercel Agent","Vercel Sandbox","AI Gateway"]
related_patterns: ["agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "Vercel adds the underlying custom-monitor capability: query-based metrics, anomaly or threshold detection, configurable filters and grouping, CLI creation, and multi-channel alerts; the prior covered Agent validation of monitor alerts before paging."
prior_refs: ["memory:dc77cb44-c915-8a22-a112-e396a3b23fde","memory:28cab208-c365-8c81-af34-6ceb8b6306ed"]
source_url: "https://vercel.com/changelog/create-custom-monitors-in-vercel-observability"
source_urls: ["https://vercel.com/changelog/create-custom-monitors-in-vercel-observability"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:vercel-observability-adds-custom-monitors-with-threshold-and-bb5dca"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/vercel-observability-adds-custom-monitors-with-threshold-and-bb5dca/","markdown":"https://newruntime.com/posts/vercel-observability-adds-custom-monitors-with-threshold-and-bb5dca.md","json":"https://newruntime.com/posts/vercel-observability-adds-custom-monitors-with-threshold-and-bb5dca.json"}
---

# Vercel Observability adds custom monitors with threshold and anomaly alerts

## Retrieval answer

Vercel Observability now lets teams create custom monitors from query results, with fixed-threshold or statistical-anomaly detection. The release matters to AI application teams because spend, sandbox consumption, workflow failures, and external API reliability can be treated as monitored operational signals rather than values someone checks manually in a dashboard.

## Summary

Vercel Observability now lets teams create custom monitors from query results, with fixed-threshold or statistical-anomaly detection. The release matters to AI application teams because spend, sandbox consumption, workflow failures, and external API reliability can be treated as monitored operational signals rather than values someone checks manually in a dashboard.

## What happened

Vercel has added custom monitors to Vercel Observability. A monitor can be created from the result of an Observability query, allowing a team to turn a charted signal into an alerting rule. Vercel describes several possible signals: AI Gateway costs, Core Web Vitals and other performance metrics, failure rates for Workflows and external APIs, and Vercel Sandbox CPU time.

Each custom monitor tracks either one metric or a ratio between two metrics. Vercel's examples are response time as a single metric and errors divided by total requests as a ratio. This distinction is important operationally: an absolute value can identify a slow response or a high cost, while a ratio can show deterioration relative to traffic. The source does not describe a prescribed set of AI-specific rules; it provides a general monitoring mechanism that can be applied to the listed signals.

For detection, users choose between a fixed threshold and statistical anomaly detection. They can also configure a detection window, filters, and grouping. The supplied material does not specify the anomaly-detection method, default window sizes, or how grouping affects notifications, so those details should be confirmed in Vercel's configuration documentation before treating a rule as equivalent to an existing paging policy.

There are several paths to creation. A user can inspect a chart's query in Observability, start from a preset, or open the team's Settings > Alerts area. Teams that want configuration in code can define a custom monitor in a JSON file and create it with `vercel monitors add`. Vercel also provides `vercel monitors schema --type custom` for the schema and examples. That makes monitor definitions capable of living alongside the application or operations configuration, rather than existing only as dashboard edits.

Alerts can be delivered through email, Slack, Vercel Inbox, push notifications, or team-level webhooks. Vercel says teams can add context for Vercel Agent to use when investigating an alert. Custom monitors are available on Pro and Enterprise plans with Observability Plus. Vercel prices the data scanned by custom monitors at $0.0035 per GB; built-in alerts remain included with Observability Plus.

## What the archive adds

This is a new fact in a known Vercel operations thread. [Vercel Sandbox now has memory observability.](https://newruntime.com/posts/vercel-sandbox-now-has-memory-observability-analysis-b8617c/) documented memory metrics as an operational signal for sandboxed execution. [Vercel Agent can now validate production alerts before they page on-call](https://newruntime.com/posts/vercel-agent-can-now-validate-production-alerts-before-51e95a/) covered a downstream investigation and paging step. Custom monitors add the configurable rule-creation layer before that investigation step.

## Trend signal

This supports [agent security moves to runtime boundaries](https://newruntime.com/patterns/agent-security-moves-to-runtime-boundaries/) only indirectly: teams operating agents need observable, bounded execution environments before they can govern them. The stronger signal here is operational rather than security-specific: Sandbox CPU time, external failures, and gateway costs are being surfaced as monitorable runtime conditions.

## Why now

The archive provides a concrete adjacent context: Vercel Agent can validate production alerts before a critical alert reaches a paging webhook. Custom monitors expand the set of query-derived conditions that can enter an alert flow and can carry context for that agent investigation. The source does not claim that the two releases form a single product workflow, but their capabilities meet at the alert boundary.

## Connections to other events

The related Field Note on [Vercel Agent validating production alerts](https://newruntime.com/posts/vercel-agent-can-now-validate-production-alerts-before-51e95a/) describes an agent checking monitor logs and metrics and forwarding only alerts it classifies as Critical to configured paging webhooks. This release does not change that validation feature. It gives teams a new way to define the monitored conditions that may produce alerts, including query-derived cost, failure, and sandbox-usage signals.

## Why it matters

AI systems introduce operating variables that are easy to see but hard to own: model-gateway spending, sandbox resource use, workflow failure rates, and dependencies outside the application boundary. A dashboard alone does not establish a response contract. Query-derived monitors let a Vercel team decide which of those values deserve a threshold, an anomaly rule, routing, and potentially an investigation context.

The CLI and JSON option is especially relevant for teams that want alert definitions reviewed and maintained with deployment configuration. It can make a monitor reproducible across environments, subject to whatever configuration-review process the team already uses. The price also means teams should evaluate monitoring cost as part of the rule design, since Vercel charges for data scanned by custom monitors. The release does not remove the need to choose useful thresholds, validate noisy alerts, or assign an owner for a response.

## Limits and uncertainty

Vercel's announcement establishes the available configuration surfaces, notification destinations, plan eligibility, and scan-based price. It does not establish how accurate statistical anomaly detection is for a particular workload, how quickly notifications arrive, what retention or query-volume constraints apply, or the total cost of a monitor in a real deployment. It also does not say that Vercel Agent automatically investigates every custom-monitor alert; it says users can add context for the agent to use when investigating an alert. Teams should test a rule against their own traffic and incident history before relying on it for paging.

## Skill to build now

Pick one bounded AI-service signal already visible in Vercel Observability, such as AI Gateway cost, Sandbox CPU time, or external API error ratio. Define the numerator and denominator if it is a ratio, choose a detection window, and write the monitor as JSON. Run it first as a routed notification rather than an immediate paging trigger, then compare its alerts with actual operational events. That exercise creates a reviewable definition of what the team considers abnormal.

## Reader takeaway

Vercel Observability can now convert its query results into configurable, code-definable monitors. For AI application teams, the practical opportunity is to put gateway cost, sandbox consumption, and dependency failures into an explicit alerting workflow—while testing noise, response ownership, and scan costs before escalating those signals to on-call.

## Open question

Will Vercel publish enough detail about anomaly behavior and scan-volume economics for teams to safely use custom monitors as paging inputs rather than dashboard notifications?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agent Security](https://newruntime.com/topics/agent-security.md): Explore the agent-security topic hub.
- topic hub: [Sandbox](https://newruntime.com/topics/sandbox.md): Explore the sandbox topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)