Nova is your AI platform engineer: it reads your clusters and applications, investigates incidents, and changes nothing your guardrails do not allow.
Ask in plain language about your clusters and applications: health, versions, resource usage, deploy history. Nova answers from what it reads through Astro Platform and your connected skills.
Each investigation belongs to a numbered incident, such as INC-42. Needs you lists what waits on a person: a fix to approve or a read to allow. The incident keeps the cause, the evidence and how it ended.
Point Alertmanager or Grafana at Nova. When an alert fires, Nova investigates and posts what it found in Nova, and in your Slack channel on the Enterprise plan.
Nova writes Helm charts, Kubernetes manifests and Terraform modules, checks them, and shows what the checks found. A file is a draft for you to review and download.
Skills decide what Nova can read and act on: Astro Platform, built-in integrations and your own MCP servers. Nova only sees what your organization connected.
astroctl nova runs the same Nova, with the same skills and guardrails, in your shell.
Your organization’s guardrails decide whether it runs, waits for a person, or is blocked. By default, status reads run; logs, changes, scaling and provisioning wait for approval; deleting and touching secrets are blocked.
Approve or decline from the card in the chat or on the incident. If nobody decides within the approval window (15 minutes by default), nothing runs.
They change the rules, try them in shadow mode first, and can turn on emergency stop, which blocks every write in the organization while reads keep working.
Guardrails activity lists each action, its outcome and who approved it.
In Nova: Settings, Guardrails. Units for each answer and investigation are in Settings, Usage & billing (Nova Units(opens in a new tab)).
Once you have access, ask Nova questions like these.
Which applications are unhealthy right now?
Investigate my unhealthiest application and tell me the likely cause
Show recent events for my unhealthiest application
Show my clusters and their health
Which Kubernetes versions can my clusters upgrade to?
Are there open vulnerability findings on my clusters?
Write a production Helm chart for a Node.js service with probes, resource limits and autoscaling
Write Kubernetes manifests for a background worker with a disruption budget
Compare EKS, GKE and AKS for a regulated workload