Debugging starts with why
Before touching anything, BashPilot reads events, logs and describe output for the affected workload. You get the actual cause, not a blind restart. Fixes come after the diagnosis, in the same conversation.
BashPilot for Kubernetes AI Kubernetes management
BashPilot puts a small agent on your control plane node and works with local kubectl access. Your kubeconfig never leaves the cluster and no SSH tunnel is involved. You ask for an outcome, it runs the right operations against the API and confirms the resulting state.
7-day free trial. Cancel anytime.
agent online
reasoning · executing · verifying
How it works on Kubernetes
From a sentence to a verified rollout
No YAML to hand-edit, no kubectl to memorize. You describe the outcome and BashPilot walks the same path a careful operator would.
1
You ask
"Scale checkout-api to 6 and check the rollout." Plain words, the way you'd tell a teammate.
2
Reads the cluster
Pulls pods, events and current replicas straight from the API before touching anything.
3
Plans the change
Chooses the exact operations from a reviewed catalog. No improvised YAML, no guesswork.
4
Pauses on risk
Draining a node or deleting a workload waits for your explicit yes first.
5
Runs on the API
Applies the change against your cluster through its scoped service account.
6
Verifies & reports
Watches until pods are Ready, then shows you exactly what changed.
What's covered
Every corner of the cluster, in plain English
Deep, native coverage of the objects you touch every day. This is a taste, not the full list.
WorkloadsDeployments · StatefulSets · DaemonSets · Pods · Jobs · CronJobs · and more
Rollouts & scalingScale replicas · Rolling updates · Rollbacks · Restarts · HPA · and more
Networking & ingressServices · Ingress · Endpoints · DNS · Network policies · and more
Kubernetes
Config & secretsConfigMaps · Secrets · Env vars · Limits & requests · RBAC · and more
Nodes & namespacesNodes · Cordon · Drain · Taints · Namespaces · Quotas · and more
Storage & observabilityPersistentVolumes · PVCs · StorageClasses · Logs · Events · Metrics · and more
Problems it solves
The 2 a.m. pages, handled
The failures every Kubernetes operator knows by heart. BashPilot reads the real evidence, fixes the cause, and confirms the recovery.
CrashLoopBackOff
Fixed
Reads the events and last logs, finds the real cause, applies the fix and watches the pod stay up.
OOMKilled
Fixed
Spots the memory limit hit, right-sizes requests and limits, then restarts the rollout cleanly.
ImagePullBackOff
Fixed
Checks the image tag, registry auth and pull secret, and clears whatever is blocking the pull.
Stuck rollout
Fixed
Inspects the rollout status and the failing pods, then unblocks it or rolls back to the last good revision.
Pending pods
Fixed
Finds why they will not schedule, from resource pressure to taints to a missing PVC, and clears it.
Node under pressure
Fixed
Reads node conditions, cordons and drains it safely, and reschedules the workloads elsewhere.
In practice
Kubernetes
Safe changes to live workloads
Rollouts are watched until they settle, and a rollback is one sentence away. Anything that could interrupt a running service, such as draining a node, pauses for your approval first.
Your kubeconfig stays put
The agent uses the cluster credentials that already exist on the control plane. Nothing is uploaded, nothing is copied out, and access ends the moment you remove the agent.
Security first
Built to be trusted with production clusters
The reason teams let BashPilot near a live cluster: it is scoped, it asks before it disrupts, and it proves every change.
Your kubeconfig never leaves
The agent runs on the control-plane node and uses the credentials already there. Nothing is uploaded, nothing is copied out.
Bounded by RBAC
It can only do what its service account permits. Grant exactly the scope you are comfortable with, and it cannot exceed it.
Approval before disruption
Draining a node, deleting a workload or scaling to zero always pauses for your explicit confirmation.
Verified, not assumed
After every change it re-reads the rollout and resource state. Done means Ready on the cluster, not a clean exit code.
Full audit trail
Every operation and its result is recorded, so you always know what changed, when, and why.
Outbound-only agent
The agent polls over TLS and opens no inbound ports. There is nothing on the cluster for a scanner to reach.
FAQ
Common questions
Which Kubernetes setups does it work with?
Any cluster where the agent can run on a control plane node with local kubectl access: k3s, kubeadm clusters, single-node labs and self-managed production setups alike.
Do I have to upload my kubeconfig?
No. That is the point of the design. The agent uses the credentials already present on your control plane and they never leave the machine.
Can it delete workloads?
Only after you approve. Destructive operations such as deleting a deployment or draining a node always show a confirmation card before anything runs.
Does it replace kubectl?
For daily work, mostly yes. You can still use kubectl whenever you like. BashPilot simply gets the routine done faster and keeps a record of what changed.
Try it on your own server.
Connect in about a minute. The first week is on us.