commands.md

Prompt & commands

The bottom panel runs commands. The retained AI mode is disabled; ctrl+a or the right-hand tag displays an unavailable notice.

• CMD ❯ — / and : commands run inside k10s; anything else runs as a shell command and its output opens in the main panel as a text view ($ date, $ kubectl get pods -o wide, $ helm list). The line is run verbatim through your own $SHELL, so aliases, functions and PATH are the ones you expect, and the CLI name in /settings is a label for hints rather than something that rewrites what you typed.

  Two limits, both deliberate: a command is given 30 seconds, and it is given no terminal — so anything interactive (vim, top, kubectl exec -it) waits and then times out. Use s on a pod for a real shell.

• AI ✦ (unavailable) — retained experimental integration, not a usable prompt mode in this build.

Search palette (ctrl+p)

One box that finds both resource kinds and individual objects. Kinds you have opened are searched by object name; kinds not yet loaded match by name only, because scanning their objects would mean starting a cluster-wide watch per kind. The footer states how many kinds are in that reduced state.

↑↓ moves, enter jumps to the kind (and row), esc closes.

Cmd+K cannot be bound to this — see keybindings.md.

Two command prefixes

• / — k10s itself. Theme, settings, updates, help. Nothing here is about the cluster you are looking at.
• : — the cluster, in the k9s vocabulary. Name a resource to open its view (:po, :deploy), switch where you are (:ns, :ctx), or act on the view you are on (:search, :filter, :scale).

Typing either prefix pre-fills it and shows only that set. ↑↓ (or the wheel) moves the highlight and enter runs the highlighted command straight away — no second trip through the prompt. The exception is a command whose argument is mandatory (:scale, :search, :filter), where enter completes it instead so the argument can be typed; a resource command's argument is optional, so enter runs it. tab always completes rather than runs.

Command	Args	Effect
/theme		theme picker with live preview
/settings		CLI name and the update check; AI controls are disabled
/mouse		toggle mouse capture — same as ctrl+s
/update	[skip]	check GitHub for a newer k10s and install it over this binary
/version		which build is running, and what the last check found
/demo		switch to k10s's built-in demo cluster (k10s-demo) — sample data; :ctx leaves it
/setup		how to install kubectl and get a ~/.kube/config — links, readable with no cluster connected
/help		keybindings + commands text view
:search	<term>	filter the resource list (left pane, by kind)
:filter	<term>	filter rows of the table currently open (main pane)
:scale	<n>	scale the selected deployment/statefulset/replicaset
:ns	[name]	the Namespaces table; with a name, switch straight to it
:ctx	[name]	context picker; with a name, switch straight to it
:aliases		every : name in one text view
:q		quit (:quit, :qa too)

Unknown commands toast unknown command … — /help lists everything.

/ns and /context are gone: namespace and context name things the cluster has, like every other : command, and :ns / :ctx is where a k9s user reaches for them. /theme still takes no name — it opens a chooser showing what is actually available.

Resource commands (:po, :deploy, …)

Every kind in the Resources pane has a : command, and answers to the same three or four spellings k9s uses — the short form, the plural, the singular. They are generated from the kind list the connected backend serves, so a backend that adds a kind gets its command for free and the names can never drift from what the pane shows.

Group	View	Names
Workloads	Pods	:po :pods :pod
	Deployments	:deploy :deployments :deployment :dp
	ReplicaSets	:rs :replicasets :replicaset
	StatefulSets	:sts :statefulsets :statefulset
	DaemonSets	:ds :daemonsets :daemonset
	Jobs	:job :jobs
	CronJobs	:cj :cronjobs :cronjob
	HPAs	:hpa :hpas :horizontalpodautoscalers
Network	Services	:svc :services :service
	Endpoints	:ep :endpoints :endpoint
	Ingresses	:ing :ingresses :ingress
	NetworkPolicies	:netpol :netpols :networkpolicies :networkpolicy
Config	ConfigMaps	:cm :configmaps :configmap
	Secrets	:sec :secrets :secret
	ResourceQuotas	:quota :quotas :resourcequotas :resourcequota
	LimitRanges	:limits :limitranges :limitrange
	PDBs	:pdb :pdbs :poddisruptionbudgets
Storage	PVCs	:pvc :pvcs :persistentvolumeclaims
	PVs	:pv :pvs :persistentvolumes :persistentvolume
	StorageClasses	:sc :storageclasses :storageclass
RBAC	ServiceAccounts	:sa :serviceaccounts :serviceaccount
	Roles	:role :roles
	RoleBindings	:rb :rolebindings :rolebinding
	ClusterRoles	:crole :clusterroles :croles :clusterrole
	ClusterRoleBindings	:crb :clusterrolebindings :crbs :clusterrolebinding
Cluster	Nodes	:no :nodes :node
	Namespaces	:ns :namespaces :namespace
	Events	:ev :events :event
Custom	CRDs	:crd :crds :customresourcedefinitions
	Custom Resources	:cr :customresources :crs :customresource

The first name is what the popup shows — the short form, the same one the sidebar and toasts use (po/api-gateway). The rest match while you type but never take a row of their own, so one kind is one row.

A bare : matches more commands than fit above the prompt, so the popup draws a screenful and scrolls — the tag counts where you are (17/37) and ↑/↓ say which way there is more. Nothing is dropped from the list; typing one more letter is still the fastest way through it.

Matching is against every spelling: :dp narrows to Deployments and then closes, because there is nothing left to disambiguate. A name that is also the start of a longer one is highlighted first, so :pv opens PVs rather than PVCs and :role opens Roles rather than RoleBindings.

The table is generated from the kind list the backend serves, so it is exactly what the Resources pane shows — nothing more. Kinds k9s has that k10s does not model (:rc, :csr, :pc, :leases, Helm releases) and its extra screens (:popeye, :xray, :pulses) have no command here, because there is no view for them to open.

The optional argument

:po                 Pods, as they are
:po kube-system     switch to that namespace and show Pods there
:po all             every namespace at once, NAMESPACE column and all
:svc api            "api" is not a namespace, so it filters the rows instead

A namespace argument only applies to namespaced kinds — :no kube-system filters Node rows rather than switching, since Nodes ignore namespaces. Arriving with no argument clears whatever row filter was left behind, so a command always shows the whole view.

Two of them are also switchers, which is where the argument saves the most time:

• :ns opens the namespace chooser (the same one the header's ns … button opens); :ns <name> switches namespace and leaves you on the view you were reading, rather than jumping to Pods.
• :ctx opens the context picker; :ctx <name> switches straight to that context — the same reconnect, without the picking. An unknown name toasts rather than reconnecting to nothing.

:aliases prints the whole table above in a text view, since four spellings per kind is more than a popup should ever list.

Growing the command box

ctrl+z grows the prompt to half the screen, and typing anything that is not a / or : command grows it automatically — a kubectl line or an AI question gets long, and a one-row field that scrolls sideways hides most of it. The value wraps across the tall box so the whole thing is readable. esc shrinks it; esc again leaves the prompt. There is a [ grow ] / [ shrink ] button in the panel's top border too.

Namespaces

The active namespace is a specific name or the sentinel all. The chooser is reachable two ways — the ns <name> ▾ button in the top-right of the banner, or :ns. Both open the Namespaces table in the main panel, whose first row is all. Pressing enter on a row switches to that namespace and jumps to its Pods, which is the usual next step.

When you already know the name, :ns <name> skips the table and switches outright, staying on the view you were reading. So does the namespace argument on any resource command — :po kube-system does both at once.

What changes when you switch:

• A specific namespace — the table shows only that namespace's rows.
• all — every row, with a NAMESPACE column prepended (accent2-colored), the same idea as kubectl get pods -A. Rows group by namespace, then name.
• Every Resources-pane badge count updates too, not just the open table.
• Non-namespaced kinds (Nodes, Namespaces, CRDs) ignore the filter entirely.
• The row cursor resets to the top, since the row set changed completely.

CLI name (/settings)

Which command you type for Kubernetes — kubectl, k8s, k, or your own — used in hints and messages. It defaults to kubectl and is never asked for: k10s opens straight into the cluster, and the status bar mentions /settings once on the very first launch rather than a dialog standing in the way.

It is a label: k10s reads the cluster through the API directly, and a line you type at the prompt is run exactly as typed, so kubectl get pods reaches whatever kubectl your PATH has, whatever this setting says. Change it any time with /settings; it is stored as cli: in config.md.

Self-update (/update)

/update asks GitHub for the newest release, confirms — same modal as delete and drain, since it replaces the binary you are running — verifies the download against the release checksums, installs it, then offers to restart into it. /version reports what is running and where updates come from without touching the network.

The same check also runs once a day at startup and only speaks up when there is something newer, as a toast plus a clickable ⇧ X.Y.Z badge in the status bar. Three levels of "not now":

esc on the dialog	not this time
/update skip	stop mentioning this release; the check stays on
/settings → UPDATES → off	stop checking at all; /update still works

Outside the TUI: k10s update and k10s --version. Full behaviour in update.md.

Read-only mode (k10s --readonly)

For looking at a cluster you must not change, such as production. The flag can go before or after demo.

• The Actions pane keeps Describe, YAML, Logs and Top. Edit, Scale, Rollout Restart, Cordon, Drain and Delete are gone, and so are Shell and Port Forward, which open a session into the cluster. Their keys, :scale included, answer with a notice instead of a dialog.
• All external plugins and typed shell commands are blocked. They run outside k10s's guarded Kubernetes transport; dangerous: false is not evidence that a command is read-only.
• The header says READ-ONLY for as long as the mode is on.
• The same rule holds on the wire: every client k10s builds refuses, before it leaves the machine, any request other than GET, HEAD or OPTIONS, the GETs that open exec, attach or port-forward streams, and anything sent through the proxy of a pod, a service or a node. A code path the UI missed still cannot write.

AI settings (/settings)

Disabled in this build. ctrl+a shows an unavailable notice and the AI settings controls cannot be used. The following retained provider/schema reference is not an active setup guide. No automatic inspection of logs, YAML, events or pod health is implemented by the provider integration.

Field	Notes
Provider	radio: OpenAI-compatible / Anthropic; switching auto-fills that provider's default Base URL + Model
Base URL	e.g. https://api.anthropic.com/v1 — inline editable
Model	e.g. claude-sonnet-5 / gpt-5 — inline editable
API Key	inline editable, always displayed masked

↑↓ moves, enter selects a radio or starts editing a field, tab reaches Save, esc closes keeping what is set. Every field is persisted as soon as you commit an edit or toggle the provider — see config.md.

The API key field never pre-fills with the stored secret; leaving it empty keeps the existing key rather than clearing it.

Requests are real (internal/ai): OpenAI-compatible posts to /chat/completions with a Bearer token, Anthropic posts to /messages with x-api-key + anthropic-version. The current context, namespace, resource kind and selected object are injected into the system prompt so answers can refer to what's on screen. Errors surface the server's own message.