On this page
Quick answer
A war room in Kepler, RubixKube's desktop app for the engineer on call, puts several AI agents on one big incident at once. Kepler leads. It starts a member for each angle, such as the pods, the deploy diff and the dashboards, shares what each one finds, and collects every change they want to make into one queue for your approval.
A big incident needs several people looking at once: someone on the pods, someone on the last deploy, someone on the dashboards, and someone keeping track. At 3am you have one person. A war room gives that person a team.
How it works
- Open a war room. Choose New war room and describe what you see. A symptom is enough, or start from Outage, Slowdown, Bad deploy or Unexplained change.
- Kepler leads. It starts a member for each angle it needs, each in its own pane, up to four side by side.
- Findings are shared. Members pin what they find to the room, so one member's finding reaches the others.
- Changes wait for you. Every change any member wants to make lands in one Waiting on you queue.
Example: checkout errors after a deploy
Checkout starts returning 502s at 09:40, five minutes after an api deploy. The pods member finds containers killed with exit code 137 against a 512Mi limit. The deploy-diff member finds image v2.14.1 raised the page size from 50 to 500. The dashboards member shows the error rate going from 0.2% to 11%. Kepler puts it together: bigger pages, more memory, killed pods. The fix, applying the previous ConfigMap, waits for your approval.
Rules of the room
- Members cannot do more than the room allows. They share the room's posture and cannot widen it.
- Only the lead starts or stops members, and closing the room stops every member.
- Some commands are always blocked, in every posture: deleting, scaling, draining and rolling back in Kubernetes, uninstalling Helm releases, destroying Terraform and removing files.
- A room runs in Assist or Yolo, because a room exists to get to a fix.
About Kepler
Kepler runs on your machine with your access, on macOS, Linux and Windows, and you can use it with or without a RubixKube account. See Kepler, or read War room.
Frequently asked questions
How many agents can a war room run?
Up to four in panes side by side. More members can work without a pane.
Can a member make changes without me?
No change runs without passing the room's posture, and dangerous commands such as delete, scale and rollback are blocked in every posture. Approvals collect in one queue.
Do I need RubixKube to use Kepler?
No. Kepler needs a Kepler sign-in and a model. It works on its own, with your local access.