KubeAgent runs inside your cluster and catches crash loops, OOM kills, 5xx spikes, bad rollouts and expiring certificates without spending AI tokens. It fixes only what you allow — enforced by the Kubernetes API server, not a prompt — and asks you in Slack for the rest. Flat price from $29.99/mo, not per node.
kubeagent login
kubeagent onboard
kubeagent agent install
The agent checks pods, nodes, and deployments with fixed rules: CrashLoopBackOff, OOMKilled, ImagePullBackOff, node pressure. It also reads your ingress 5xx rate from Prometheus and compares it before and after each rollout. Healthy checks and logs never reach the AI, so watching costs zero tokens.
When an incident is new or has changed, KubeAgent fetches logs, events, and your cluster's knowledge base, and traces the root cause with read-only kubectl lookups. An incident it already diagnosed is not diagnosed again until a cooldown passes, even after the agent restarts.
If the fix is on your allow-list, the agent applies it and tells you what it did. If not, you get an Approve/Deny request. A bad rollout can be rolled back to the previous revision, but only after you approve.
One agent per cluster. Works on any distribution kubectl can reach.
Read the docs →Pick the actions in onboarding: restart_pod, rollout_restart, scale_deployment, set_resources. After an OOM kill the agent can raise memory, never lower it, up to 4x the starting value and 16Gi, once per deployment.
When a fix is outside your allow-list, the owner gets a Slack button that opens a dashboard page. Each link is owner-only, works once, expires after 24 hours, and is audit-logged.
Ingress 5xx rate and post-deploy regression checks are on by default and use zero AI tokens. Alertmanager and webhook alerts can be sent in as an option.
Each issue gets a fingerprint. An unchanged issue is not diagnosed again until a cooldown (6 hours by default) passes. State survives agent restarts.
Slack, Discord, Microsoft Teams, Telegram, PagerDuty, or custom webhooks for incidents and recovery updates.
Onboard once and KubeAgent learns your services and stack. That context goes into every diagnosis, and past incidents feed back into it.
The dashboard shows a banner and sends one email when your agent is behind. Update with kubeagent agent install --upgrade, which keeps your settings.
Connect once in the dashboard. KubeAgent sends incident alerts straight to your team's channel — no polling, no extra tab. Slack can also trigger an AI diagnosis with one click.
One-click AI diagnosis
Server webhooks
Channel connector
Bot notifications
Incident escalation
Send anywhere else
Compare: Claude Code + Kubernetes MCP · Grafana Assistant · HolmesGPT · Metoro · all comparisons · why we built it this way
KubeAgent never asks for your kubeconfig. The agent runs inside your cluster on its own service account, or as a CLI with your local kubectl context. It only makes outbound calls.
╭──────────────────────────╮
│ YOUR NETWORK │
│ kubeconfig · kubectl │
│ │
│ ┌────────────────┐ │
│ │ kubeagent agent│ │
│ └────────┬───────┘ │
│ │ │
╰────────────┼─────────────╯
│ OUTBOUND ONLY
▼
┌───────────────┐
│ api.kubeagent │
│ alert channels│
└───────────────┘
✓ NO inbound cluster access
✓ NO stored credentials
✓ NO long-lived tokens
Monitoring and alerts are free forever. Every account starts with 200K AI credits for diagnosis and auto-remediation — they never expire. No credit card required. All plans include Slack, Discord, Teams, Telegram & PagerDuty integration and full dashboard access.
kubeconfig. Our servers never connect to your cluster. Your credentials never leave your machine or network. AI requests are routed through our servers for billing, but we do not log or store the content. Nothing is retained unless you explicitly opt in for edge-case debugging.kubeagent login, then kubeagent onboard to scan your cluster, build a knowledge base, and choose your auto-fix allow-list. Then run kubeagent agent install for the always-on agent, or kubeagent watch to monitor from your terminal. About 5 minutes.restart_pod, rollout_restart, scale_deployment and set_resources. The default is the first two. A memory or CPU raise can only go up, by at most 4x and to at most 16Gi memory or 8 CPU, once per deployment. Everything else is refused and sent to you as an Approve/Deny request.kubeagent agent install --upgrade to update it and keep its settings.