From visibility to autonomy.

Kobi Copilot is Kobi's assisted mode, already working in your cluster: it investigates the incident with real context (events, logs, metrics), proposes the diagnosis and executes actions under your supervision. Every action is audited — who, what and why. It doesn't replace your judgment; it saves you the slow part of the investigation.

• → Kobi Copilot — assisted incident resolution with audited actions
• → kubectl operations, Helm visibility and a read-only MCP server, via agent

The jump from fixing the symptom to fixing the cause: Kobi locates the workload's source manifest and proposes the fix as a pull request in your repo, with the rationale and the before/after in the body. The merge is the approval; your GitOps or your pipeline applies it. And deterministic Skills precompute frequent diagnostics: instant answers, fewer credits.

• → Root-cause resolution — the fix corrects the manifest and lands as a pull request in your Git repo; the merge is the approval
• → Deterministic Skills — faster answers, fewer credits

Autopilot detects, investigates and opens the remediation pull request on its own, with the incident evidence in the PR body. Policy decides how much autonomy: from propose-only to auto-merge when your checks are green, always respecting your branch protection.

• → Remediation PRs opened autonomously, with the incident evidence included

Autonomous remediation debuts as an open beta on every Cloud plan: you enable it per organization, define the time window it can act in and the mode (suggest-only or execute). Destructive actions require human approval by platform policy, and its usage is metered separately against the same plan credit pool Kobi draws from.

• → Autopilot in open beta — org-level opt-in, time window, guardrails, approval queue and automatic postmortem

The final form: Autopilot operates your whole fleet under a policy you define, with an approval queue for the sensitive and an automatic postmortem for every intervention. You govern the policy; it runs the day-to-day.

• → Cross-cluster Autopilot — one incident spanning clusters, resolved on a single decision thread

24 rules that evaluate reliability and configuration from minute one, without writing a single one: missing probes, resource limits, single replicas, unpinned images. Findings feed Kobi, so the copilot already knows the weak spots when an incident hits.

• → Insights Engine — 24 reliability and configuration rules, zero config

Your security tools already produce findings; the problem is they live in four places. KubeBolt detects them (if Trivy, Falco or Kyverno run in your cluster, it finds them on its own), normalizes their findings and presents them in a single dashboard: vulnerabilities, runtime threats, CIS benchmark and policy violations, prioritized.

• → Security & Compliance dashboard — Trivy, Falco, CIS and Kyverno in a single view

Ask Kobi about your security posture like you'd ask a teammate: what's the CIS score, which threats are active, which policies are being violated. Answers come from your real findings, not generic documentation.

• → Kobi over your findings — CIS score, active threats, policy violations

Security Skills start running on their own under policy: CVE triage by real exposure, runtime threat response and policy drift correction. Your team reviews the exceptional, not the routine.

• → Autonomous security Skills and autonomous FinOps under policy

Pulled forward from stage 02: the Cost tab appears on every cluster once OpenCost data flows, without building your own FinOps stack. Monthly run-rate, idle cost, breakdown by namespace, workload and node, and money-priced rightsizing recommendations from your real consumption — with three connection modes in the wizard.

• → Cloud costs on OpenCost — per-cluster spend, idle and money-priced rightsizing

Visibility already lives in the Free plan (the Cost tab on OpenCost); this stage deepens it: detection of orphaned volumes and abandoned resources still running up the bill, so the Explorer shows not just the spend but what you can delete painlessly.

• → Actionable costs — detection of orphaned volumes and resources still billing you In progress

The rightsizing recommendations already exist, priced in money; here they get executed: propose, approve, execute. Alongside scheduled shutdown of the environments nobody uses at night, with the achieved savings measured in an ROI dashboard.

• → From math to action — rightsizing applied with approval, scheduled non-prod shutdown and an ROI dashboard

Stop hopping between kubectl contexts: every cluster in one panel, with a global search that finds a pod, a deployment or a configmap wherever it lives. And if you run ArgoCD or Flux, KubeBolt reads your apps and maps what deploys from which repo, with its sync and health.

• → Multi-cluster fleet overview and global cross-cluster search
• → GitOps topology — ArgoCD apps with sync and health today; repo mapping and Flux still building In progress

A single cockpit for AWS, GCP and Azure: findings, cost and insights aggregated cross-cluster, and cost reconciled against each cloud's real billing.

• → Multi-cloud cockpit — AWS, GCP and Azure with reconciled cost in one place