Skip to main content
Watcher runs on your cluster and watches for crashes, failing health checks and high CPU, memory or disk use. When something trips, it writes a diagnosis and can send it to Slack. Watcher is separate from recovery. The orchestrator already restarts and reschedules managed workloads within its policies. Watcher tells you what happened and why. Any fix it proposes needs your approval.

​
Before you start

  • Your app is deployed to a cluster. See Deploy an existing app.
  • For Slack alerts, a Slack workspace where you can add an app to a channel.

​
Ask Monk

The defaults suit most clusters. Change them in your request if you need to:

​
What you review

  1. Slack. If you want Slack alerts and Monk has no Slack connection yet, it asks you to authorize Slack and pick a channel. The webhook is stored like any other credential. You can also enter an incoming webhook URL by hand in the local dashboard. Without Slack, Watcher still runs. You just don’t get notifications.
  2. The plan. It shows the thresholds, where Watcher runs and the tokens it creates for itself. Nothing is deployed until you approve.

​
Default thresholds

CPU and memory alerts fire when usage stays above the threshold for 5 minutes.

​
What you get

Slack messages with what happened, the log lines around it and a written diagnosis. By default only these AI-written alerts go to Slack. To act on an alert, open your coding agent and ask Monk about it:
Any change follows the usual plan and approval. Slack is for notifications only. You can’t chat with Monk or approve anything from Slack.

​
Check or remove Watcher

Removing Watcher opens an approval.

​
Next