Journey overview on Grafana Labs

Journey overview

Tue, 20 Oct 2026 13:08:30 -0400

Keep watch on a service without writing every rule first

Describe in plain language what a service should look like, including the noise you expect, and a watcher checks it on a schedule. This journey explains what Assistant Watchers are and when to use one, then walks you through creating, calibrating, running, and reviewing a watcher in your Grafana Cloud stack.

What you’ll learn

Here’s what this journey covers:

What a watcher is: What it checks and the assessments it reports.
When to use one: The kinds of monitoring work a watcher suits.
How it differs from an alert rule: You describe intent instead of writing a condition, and Assistant can recommend either or both for a service.
How to create one: Keep the scope narrow, give context Assistant can’t infer from telemetry, and let Assistant build the checks from your data.
How to read a result: Trigger a run, check the standing verdict, and review the evidence behind it.

Journey duration

~20-30 minutes

What's a watcher?

Tue, 20 Oct 2026 13:08:30 -0400

What’s a watcher?

A watcher checks your telemetry on a schedule and reports notable changes. You describe what to monitor, Assistant builds checks from your data, and you review the findings and adjust the watcher as your service changes. Depending on the data sources you select, a watcher can check metrics, logs, trace-derived metrics, profiles, and other data.

Why use a watcher?

Service health checks: A watcher checks the same scope on each scheduled run and notifies you when an issue needs attention, not on every run.
Services with expected noise: Describe the behavior you expect, such as brief error spikes during deployments, and focus the watcher on the signals that need attention.
Services that change over time: Turn on automatic recalibration, and Assistant reviews and updates the watcher’s checks on a schedule.
Follow-up monitoring after an incident: Keep a watcher running on the affected service, and set a stop time for a limited monitoring window.

What each assessment means

The Watchers list shows each watcher’s latest assessment or check status:

Label	Meaning
All clear	The latest completed run found no concerns.
Flagged	The latest completed run found a warning-level concern.
Escalated	The latest completed run found a critical concern.
No verdict yet	No reliable assessment is available yet.
Checks failing	The latest finished run failed. Open Run history to diagnose it.

A watcher can also notify Slack, Microsoft Teams (if available on your stack), or a webhook endpoint, send findings to Grafana Alerting, and launch an Assistant Investigation for a critical finding. The hands-on path leaves those actions off so you can create, calibrate, run, and read a result first.

How a watcher differs from an alert rule

Tue, 20 Oct 2026 13:08:30 -0400

How a watcher differs from an alert rule

A Grafana alert rule is a query you write with a fixed condition. A watcher is different: you describe what healthy looks like, and Assistant turns that into checks against your data.

For checkout, an alert rule fires when error rate stays above 5% for the pending period you set, which is how long the condition must be met before the rule starts firing. A watcher can also use context you give it, such as that short spikes during a rollout are expected and that checkout should recover within a few minutes.

	Alert rule	Watcher
You provide	A query, condition, and pending period	Intent: what healthy looks like, including noise you expect
What “bad” is	A threshold you wrote	Checks Assistant derived from your data
What it produces	Fires when the condition holds for the pending period	An assessment for each completed run: All clear, Flagged, or Escalated
How it notifies	To a destination you choose, directly or through routing rules	Through the Slack, Microsoft Teams, webhook, or Grafana Alerting notifications you turn on

An alert rule’s destination is a contact point, for example an email address or a Slack channel. The routing rules that decide which alert reaches which contact point are notification policies.

How they work together

A watcher can send its findings to Grafana Alerting. Your notification policies and contact points then control delivery, the same way they do for alert rules.

To decide for a specific service, use Assistant’s Alerts & Watchers mode. Describe the service or scope you want to review, and Assistant recommends alerts, watchers, both, or no notification based on your telemetry and needs.

The hands-on path creates a watcher and leaves its notification and investigation actions off.

Prerequisites

Tue, 20 Oct 2026 13:08:30 -0400

Prerequisites

To complete the hands-on path, you need the following.

Requirement	Why you need it
Grafana Cloud account (Create an account)	Watchers need a Grafana Cloud stack.
Grafana Assistant enabled	Watchers are an Assistant capability. Accept any required Assistant terms.
A supported data source with telemetry for a real service	Watchers support Prometheus, Amazon Managed Service for Prometheus, Loki, BigQuery, Infinity, CloudWatch, Synthetic Monitoring, Tempo, and Pyroscope data sources that you can read and query. Assistant builds checks from live telemetry, and queries that return no data can stop calibration. Watcher runs use the stack’s Assistant service account, not your identity. Data sources with team label-based access control (LBAC) rules aren’t supported.
Watchers permissions	`grafana-assistant-app.watcher-agents:read` to view, `:create` and `:write` together to create a watcher (creating also starts calibration), `:write` to calibrate, start, and pause, `:create` to run once, and `:delete` to delete. You also need `plugins.app:access` for the Assistant plugin.

Complete setup first

If Assistant isn’t enabled on your stack, complete Get started with Grafana Assistant first.

For the full permission list, refer to Manage Assistant access with RBAC.