Each story follows a single real-world situation from the first signal to the merge request on your desk: what the assistant looks at, how it reasons, which checks it runs before you ever see the change, and what the MR says. Examples are illustrative and use fictional names; the mechanics are the ones the assistant runs.
Read every changelog between here and the newest release, check which breaking changes touch your values, and stop at the right boundary.
Read the story →Thirty days of real usage, classified and clamped, turned into a request/limit change with the numbers in the MR.
Read the story →Hand-applied workloads and Tiller-era releases brought into git from their live state, so the first reconcile changes nothing.
Read the story →Drop-in v2 replacements with values verbatim. Migrate first, delete second, never both in one MR.
Read the story →Missing requests, security-context gaps, drifted storage classes: one small, clean MR per finding.
Read the story →Renames grounded in what each file actually contains, never guessed from its neighbours.
Read the story →From an unscraped metrics port to a ServiceMonitor, then a Grafana dashboard whose every panel is proven to return data.
Read the story →An alert fires. The assistant investigates read-only, finds the cause, and asks whether a manifest change would have prevented it.
Read the story →Spot nodes get reclaimed. Disruption budgets, topology spread and replicas keep the service up through it.
Read the story →Fifty warnings, one root cause, one issue, closed automatically when it goes quiet.
Read the story →Failed Velero and CNPG backups grouped into one issue with a diagnosis, before you need a restore.
Read the story →Idle volumes, hibernated databases and forgotten environments, with the evidence.
Read the story →Denials in the audit log clustered, classified, and turned into a narrowly scoped policy change, or an issue when it is not that simple.
Read the story →Committed Secrets become ExternalSecrets backed by Vault; broken references are caught before a restart.
Read the story →Control-mapped posture reports, image provenance, forge, cloud and edge posture. Findings, not surprise changes.
Read the story →Policies, auth methods and roles in git with a CI apply, linted for escalation risks and dangling references before they merge.
Read the story →Maintenance windows declared in git with an end time, applied by your CI, linted and audited by the assistant.
Read the story →Every record checked against live ingresses and load balancers; dangling records removed in the same wave as the retirement.
Read the story →Playbooks and inventories as first-class targets: upgrades, hygiene and failure triage from the CI job log.
Read the story →NetBox, the network, Ansible, Kubernetes and cloud KMS joined into one picture, so the root cause is found where it actually is.
Read the story →Ask about a workload in a thread and get a draft MR back, or tell the assistant to leave something alone.
Read the story →Several reviewer personas on your own MRs, each naming its model, plus an independent second opinion on every bot MR.
Read the story →A POLICY.md or a skip annotation, and every agent respects it.
Read the story →Comments on MRs are answered and change how the agents behave on your estate.
Read the story →Draft MRs on your forge, and cluster access through the GitLab agent (KAS) with no direct path to the API server.
Read the story →A sealed, read-only assessment of a new estate that physically cannot emit anything, then full support.
Read the story →Your own GPUs for sensitive work, masked identifiers for cloud models, and per-task routing you choose.
Read the story →