Assistant Watchers are now generally available in Grafana Cloud
Assistant Watchers are now generally available in Grafana Cloud
Assistant Watchers are now generally available. Watchers are always-on agents for the problems that are hard to write an alert rule for. You can’t always spell out every way something might go wrong, but you know what right looks like.
Describe it and pick your data sources. Assistant calibrates checks against your data in the background, so you can move on while it works. Once it’s ready, the watcher runs on a schedule and tells you when right starts drifting into wrong.
For the public preview announcement, refer to Introducing Assistant Watchers: agentic alerting for Grafana Cloud. Since then, Watchers can cover more of your stack and need less upkeep:
- Watch more of your stack: Prometheus, Amazon Managed Service for Prometheus, Loki, Tempo, Pyroscope, BigQuery, CloudWatch metrics, and Synthetic Monitoring results, plus Grafana-managed alert checks. Watchers can also read external HTTP APIs through Infinity, like a provider’s status page.
- Stay tuned automatically: with auto-recalibration, watchers adjust their checks and baselines as your system changes, so they stay accurate without you stepping in. When you want to steer one yourself, recalibrate straight from a Slack notification.
- Send findings where your team works: post to Slack, Microsoft Teams, or any webhook. Critical findings can also launch an Assistant Investigation automatically.
- Get a second opinion on your alerting: Alerts & Watchers mode in Assistant reviews your current alert rules and watchers, and recommends an alert, a watcher, or both for each failure mode.
- Watch for a set window: set a watcher to stop automatically after a rollout or a post-incident follow-up.
Note
Starting December 1, 2026, Watchers are billed on Assistant token usage, the same as Assistant and Investigations, with no per-watcher or per-run fee.
Go to AI > Watchers in the navigation and create your first watcher.
