Changelog

The first stable release. The app is the same as 0.10.0; from here on, a change that breaks how it is used waits for 2.0.0.

Added

• Settings › Appearance has a Text size picker that makes all the text in the app smaller or larger without zooming the rest of the window.

Fixed

• Turning a toggle on or off, such as the log console's Live or its wrap and timestamp switches, no longer nudges the buttons beside it sideways.
• Opening a link to an object in another namespace than the one selected selects the object's namespace, so the namespace selector and the list you go back to match the page; under All namespaces the selection stays.

Added

• The log console has a Copy logs button that copies every line of the buffer, and ⌘A or Ctrl+A in the console selects the log alone, a copy of it carrying the whole buffer too.
• A workload's log console can hide the pod names at the start of its lines.
• A detail page's row of tabs fades at an edge with more tabs past it, under an arrow that scrolls the row that way.

Changed

• A detail page's tabs run in one row under its header instead of down a rail on its left, so its content takes the page's full width.
• The log console's line display toggles, Wrap long lines, Show pod names and Show timestamps, are one button group, apart from the search's own toggles.
• The small buttons on toolbars and panels, the log console's among them, are a size larger.
• A deployment's image is a row of a Details card on its Overview instead of unlabelled text under its name.
• The log console's display options, Wrap long lines, Show timestamps and Tail, are on its toolbar instead of behind a View button, and on a narrow window the toolbar drops its labels and moves the search to a second row rather than running off the edge.
• Log lines carry no line numbers, and a workload's lines are labelled with the part of the pod name that tells its pods apart, in square brackets, rather than the whole name.
• Log timestamps are shown in your own time zone to the millisecond, with the offset from UTC.
• A wrapped log line runs on under its pod label and timestamp instead of in a column beside them.

Fixed

• A long object name no longer breaks the top bar: it is cut in the middle, and Restart to update stays on one line.
• On a narrow window the top bar no longer draws its breadcrumb under Create resource or pushes its controls past the window's edge: it drops crumbs, then words, as it narrows.
• On a narrow window a detail page's actions move under its name instead of covering it.
• A detail page's metric cards share the row between them instead of leaving part of it empty.

Changed

• Settings is split into sections down a rail on its left, General, Appearance, Cluster data, Connection, Charts, Updates and About, and a screen that points at a setting opens the section it is in.
• Quick actions is a lightning bolt beside the app's name and Settings a gear at the end of the top bar, each showing its keys in a tooltip, rather than rows at the foot of the sidebar.

Fixed

• A custom resource definition whose storage version is no longer served, as during a version migration, lists and opens its instances through a served version instead of showing them as not found.
• A Job, a PersistentVolumeClaim or a PersistentVolume exported cleaned for another cluster applies there, without the selector, volume binding or node the original cluster generated for it.
• A CronJob's Run now pressed twice in the same minute starts a second run instead of failing as already existing, and a Job's Run again keeps waiting for the old run to go when a read fails on the way, instead of creating over it.
• A log search written as a regular expression is used exactly as typed, so a leading or trailing space no longer changes what it matches, and highlighting marks every match of a pattern that can also match nothing, such as \d*.
• A port forward that could not bind its local port starts again when Start or the top bar's restore is pressed once the port is free, and the top bar offers every remembered forward that is not running, including a Service or another port sharing a running pod's name.
• A scheduled update check that fails while offline no longer shows Update failed in the top bar for a moment.
• A certificate the app does not trust because its issuer is unknown or it is not valid yet, such as one a TLS-inspecting proxy presents, and a proxy answering HTTPS in plain text read as Cluster unreachable with its hints instead of an unknown "fetch failed" error.
• A chart values schema whose key patterns escape punctuation the way Go allows, such as \_ or \:, no longer marks every key those patterns cover as not allowed.
• Upgrading a release uninstalled with its history kept says to uninstall it without keeping its history before installing it again, instead of pointing at an install that is refused as well.
• Usage charts and pod and node usage keep updating when a metrics read hangs on a cluster that never answers or a credential plugin waiting for a sign-in, and after a context switch they no longer start with a point from the cluster just left.
• A Helm install or upgrade whose hook the cluster refuses to create leaves a release the Helm CLI can still list, read and uninstall, and a rollback after a failed upgrade no longer leaves two revisions marked deployed.
• A chart whose Chart.yaml writes its version like a number, such as 1.10, can be installed and upgraded instead of being refused as holding another version.
• A detail tab whose read fails says why and offers Retry instead of showing an empty tab, "0 pods" or "Nothing else names this object"; a failed log download, clipboard copy or context or namespace switch shows an error, and a missing node, namespace or cluster-scoped custom resource no longer reads as missing from the current namespace.
• A pod whose init container crash-loops or cannot pull its image reads CrashLoop or Error and raises that alert instead of Pending, the ready count includes native sidecars as kubectl's does, a suspended Job reads Suspended instead of Running, and a volume snapshot the controller failed reads Failed instead of Pending.
• A crash-loop alert found from back-off events gives the span its count covers, so a pod that has been backing off for days no longer reads as hundreds of back-offs in the last 10 minutes.
• An event that keeps repeating, such as the scheduler's FailedScheduling, sorts and shows its time by when it last happened rather than when it first did.
• Saving a custom resource's manifest shows the saved text on its Manifest tab and the new values on its screens at once, instead of the text from before the save until the page was reopened.
• A pod owned by a custom resource that shares a built-in kind's name, such as an OpenKruise StatefulSet or a Volcano Job, no longer links its owner to the built-in kind's screen, and its Restart no longer rolls the built-in workload of the same name.
• The Drain dialog opens on a fresh plan with Drain on offer after a drain finished, was stopped or was refused, instead of showing the last drain's progress until the node's page was left.
• Opening another object of the same kind from a link, Back or the palette starts its screen afresh, so a revealed Secret value, an edited manifest, a drain's progress or a log search no longer carries over from the object before.
• A port forward closes a connection on your machine when the pod side closes it, so a request after a server's idle timeout no longer hangs, and a connection to a pod port nothing listens on is reset with the cluster's reason shown on the forward instead of hanging.
• Installing, upgrading and rolling back a Helm release reads the chart's objects the way Helm does, so a file mode written 0644 is octal instead of being refused, and 0400 no longer applies the wrong permissions.
• A file dropped on the window while the app is still starting, or after it failed to load, no longer replaces the app with that file.
• Switching context or namespace with an unsaved manifest edit and choosing Keep editing cancels the switch instead of running it on the next navigation, and moving to another tab of the same object keeps the edit without asking to discard it.
• A live list whose credentials can no longer be renewed, such as after an SSO session ends, says the watch failed and keeps retrying, instead of showing its last rows as live without a word.
• A custom resource named like a built-in kind, such as Longhorn's Node or OpenShift's Ingress, is created and saved in the scope its own definition gives it rather than the built-in kind's, so it no longer lands in the default namespace or is refused from the editor.
• A namespace's Overview leaves finished pods, such as completed Job pods, out of what its pods requested, so usage is read against what the scheduler still holds for them.
• Switching a chart's version after editing its values keeps your changes and takes the new version's defaults for the rest, instead of installing the old version's defaults as values you set.
• Changing the proxy, the bypass list or the CA bundle keeps the context and namespace the app is on instead of switching back to the one it opened on.
• Upgrading or rolling back a Helm release across a chart that moves an object to a new API version, such as an autoscaler from autoscaling/v1 to v2, keeps that object instead of deleting it as one the new revision no longer renders.
• Logs and shells show accented letters, CJK characters and emoji intact when one arrives split across two chunks, and a downloaded log cut to its size limit no longer drops its first whole line or opens mid-character.
• Reloading the window, closing it, a renderer crash or a context switch while a shell, a port forward, a log follow or a drain is still connecting now ends it, so no exec session stays open and no local port stays bound, and a forward or follow cut short by a context switch says why instead of going quiet.
• Memory and CPU amounts written with a lower-case k or an m suffix, as the API server often stores them, are read at their real size, so limits, namespace requests and quota usage no longer come out a thousand times too small or too large.
• A pod's Related tab lists the NetworkPolicies with an empty pod selector, which apply to every pod, and no longer lists one whose matchExpressions exclude the pod; the Network Policies and Pod Disruption Budgets lists print a selector's expressions and tell a selector matching every pod from one matching none.
• A shell whose connection drops ends and says so instead of staying open with every keystroke lost, and a shell that fails to start, such as on an image with no /bin/sh, prints the cluster's reason.
• A node's Describe tab shows the events the kubelet records about it, such as NodeReady, NodeNotReady, Rebooted and the pressure conditions.
• The Manifest tab follows the object after every refresh or scale instead of stopping at the first, and a change made elsewhere while editing no longer counts as an unsaved edit of yours.
• Add repository and New namespace stay open until the cluster or the source answers, so a refused URL or a name already taken leaves what you typed there to correct instead of closing on an error toast.
• A deployment's Compare revisions opens on the newest two revisions when the page is reloaded or opened from a link, rather than showing empty pickers or "No differences".
• Choosing the context already in use from the command palette no longer reconnects to it, so port forwards, log follows and shells keep running and the open page stays where it is.
• A log follow the cluster refuses, such as one on a container still starting, shows the cluster's reason instead of a dump of the response headers.
• A live list no longer loses a change made while it was still loading or reloading, so a pod deleted in that moment no longer keeps its row and a changed one no longer keeps its old status.
• A NO_PROXY or Never proxy these hosts entry written as *.corp.example or as a bare IPv6 address sends matching clusters direct, as kubectl does, instead of through the proxy.
• An OCI registry whose token server is on plaintext http is asked for a token without the registry's password, instead of being sent it in the clear.
• Saving settings keeps a symlinked settings file linked even when its target does not exist yet, keeps the file's permissions, and deleting the settings file no longer brings back preferences from before it moved.

Fixed

• Installing a chart hands Helm only the values you changed, so a default the chart writes as yes reaches the release as true, as helm install reads it, and the release records only your values.
• Rolling back a Helm release applies its objects server-side, as Helm does, instead of replacing them whole, so replicas an autoscaler or kubectl scale set and annotations other tools added survive the rollback, and a field another manager owns refuses it before anything is written.
• Upgrading a Helm release the Helm CLI wrote client-side is no longer refused over Helm's own fields, and a field another manager owns is named with its manager rather than shown as the API server's raw answer.

Added

• A chart version from a configured repository or registry can be installed as a Helm release: the values are edited against the chart's own schema, a review lists every rendered object with a server-side dry run and flags a chart that calls lookup, and the install follows Helm's order of CRDs, hooks and objects, recording a release the Helm CLI reads, or a failed one that can be uninstalled. Rendering needs Helm 3 or later on the machine. An install starts from Install on a chart under Helm › Charts, from a repository's row under Settings › Charts, or from Install a chart… in the command palette, and each says Helm is required, linking its install page, when none is found.
• A Helm release can be upgraded to a chart version from a configured repository or registry, from Upgrade in its header: the values start from the current revision's own, and a review shows them and every rendered object diffed against what the cluster holds, through a server-side dry run, with what the new revision removes, before the upgrade writes a new revision in Helm's order of hooks, objects and deletions and marks the previous one superseded. Objects are applied server-side as Helm 4 applies them, so replicas an autoscaler set and other fields another manager owns survive, and a field the chart would take from another manager is refused, naming both.

Changed

• Helm has its own section at the bottom of the sidebar, with Releases and then Charts, reached with ⌘7 or Ctrl+7; Add-ons keeps the CRDs, admission and API server screens, and links to a release or the charts under the old Add-ons paths no longer open.
• Helm › Charts lists the charts the configured classic repositories publish, each with Install, instead of the charts the installed releases came from; with no repository configured it says so and points at Settings › Charts.

Fixed

• A Helm release rolled back from the app keeps Helm's ownership label and annotations on every object it applies, so a later helm uninstall with Helm 4.3 or newer removes those objects instead of leaving them behind as not owned by the release.

Added

• A Deployment's and a StatefulSet's detail page has Scale in its header, which sets the replica count to an exact number.
• An .rpm package for Fedora, RHEL and openSUSE is published for x86-64 and arm64 beside the AppImage and the .deb, and like the .deb it is updated through the package manager.
• A Helm release opens on a Resources tab listing every object its current revision rendered, read live from the cluster with each one's own status, under a health roll-up that names the object making the release unhealthy, such as one that is missing.
• Linux builds for arm64, as an AppImage and a .deb, which update the same way as the x64 ones.
• A Skip to content button as the first Tab stop moves focus past the sidebar and top bar, and whatever has keyboard focus is outlined in the theme's accent colour.
• A screen reader hears the state of a shell session, a port forward, a drain and a log console's follow as it changes, and a status dot that stands alone is read by what it means.

Changed

• The Scale column in the Deployments and StatefulSets lists joins the steppers and the count into one group, and the exact number is set from a button beside it that appears on the hovered row rather than by clicking the count.
• When the operating system asks for reduced motion, dialogs and menus appear without transitions, spinners and the Live dot stand still, and the shell's cursor stops blinking.
• Copy link is back in a detail page's header as a labelled button, the first of its actions, instead of an icon beside the breadcrumbs.

Fixed

• A detail page that cannot be read says why — no permission to see the object, credentials the cluster rejected, a cluster that did not answer or took too long — instead of only "Failed to load", and a custom resource or Helm release that has been deleted reads as not found.
• An object deleted from its own page no longer stays listed after you land back on its list, which happened when the list had still been refreshing as you left it.

Added

• Keyboard shortcuts for moving between the six domains (⌘1 to ⌘6, or Ctrl+1 to Ctrl+6), going back and forward, focusing the screen's search box with /, and opening Settings with ⌘, or Ctrl+, on every platform, with a cheat sheet on ?, in Help › Keyboard Shortcuts and in the command palette.
• Copy link, beside the breadcrumbs on every detail page, copies a kubermeister:// link that names the cluster by its API server, so it opens the same page and tab for anybody whose kubeconfig reaches that cluster, whatever their context for it is called. Opening one asks before it switches context and never does more than open the page.

Changed

• ⌘R and Ctrl+R refresh the screen's data instead of reloading the window, which ended every port forward, shell and log follow; View › Force Reload still reloads it.
• A detail page keeps its tab through a reload, and returning to an object with Back opens the tab you left it on, rather than Overview.
• Starting Kubermeister while it is already running brings the open window forward instead of starting a second copy.

Fixed

• The manifest editor checks and completes against the cluster's schema on a real cluster. In 0.6.1 every schema lookup failed, so the editor quietly stayed plain YAML.

Added

• The manifest editor and the Create screen check a manifest as you type against the schema the cluster publishes for its kind, custom resources included: a wrong type, a value a field does not allow and a missing required field are marked where they are, and a field the kind does not have is flagged with the one it was probably meant to be.
• The manifest editor completes field names and allowed values from the same schema, required fields first, and shows a field's type and description when you hover its name.

Changed

• Settings are kept in ~/.config/kubermeister/settings.json on every operating system, a file you can write by hand, keep with your dotfiles and provision onto a machine: every key is optional, a change made in the app rewrites only that key, and the settings you already have are carried over on the first launch.
• A settings file that is not valid JSON, or holds a value Kubermeister refuses, is no longer reset to the defaults and overwritten: Settings names every value it did not use and why, and a file it cannot read is left untouched while the top bar says so.
• An edit to settings.json takes effect while Kubermeister runs, within a second of being saved, the same way a change made in Settings does.
• A published JSON Schema for settings.json gives editors completion, a description and the default of every key, and flags a value Kubermeister would refuse; a file the app creates points at it.

Fixed

• A pod's Logs and Shell tabs open on the container running the app instead of failing on an init container listed ahead of it. Logs can also read an init or ephemeral container, which the picker marks as such, and Shell no longer offers an init container, which has no process left to attach to.
• PersistentVolumes, StorageClasses, ClusterRoles and ClusterRoleBindings can no longer be deleted in bulk from their lists, where a plain confirmation skipped the typed name their single delete asks for. Their rows can still be checked and exported.
• A port forward started from a Service reaches its pods on the Service port's target port, where it went to the Service's own port number and failed whenever the two differed.
• Saving an autoscaler's bounds keeps every metric it watches. Changing only the replica range no longer rewrites its metrics, and changing the CPU target replaces that target alone, where either used to leave an autoscaler that also watched memory or a custom metric watching CPU only.
• A pod's shell knows the size of its terminal: it opens at the panel's size and follows it as the window or the layout changes, so full-screen programs such as top and vi draw for the space they have. The terminal used to tell the pod no size at all.

Added

• Quitting asks first, naming what it ends: every port forward, shell session, log follow and drain in progress goes with the app, and none of them come back. The dialog's Don't ask again and Settings › General turn it off.
• The rows checked on a list save to one YAML file, either as the cluster holds them or stripped of the fields the server owns so they can be applied to another cluster.
• The Create screen opens a manifest from a file, either through Import or by dropping the file anywhere on the window. The file is read by Kubermeister itself and lands in the editor, where it is applied the same way as one typed in.
• The Create screen starts a Job, a PersistentVolumeClaim, a ServiceAccount, a Role, a RoleBinding, a NetworkPolicy or a HorizontalPodAutoscaler, alongside the kinds it already offered.
• A Helm release's detail screen shows the objects a revision rendered, beside the values it was installed with.
• Help › Report a Bug… and a link on Settings › About open the bug report form with the version and the operating system already filled in.
• The update popover and Settings › About name the download size while an update comes down, so a differential update shows the few megabytes it is actually fetching rather than a bare percentage.
• Every release carries a SHA256SUMS file, so a download can be checked before it is run.
• Cluster calls go through your proxy. HTTPS_PROXY, HTTP_PROXY and NO_PROXY are read the way kubectl reads them, from your login shell as well as the window's own environment, so an app started from the Dock reaches what a terminal reaches; Settings › Connection can name a proxy instead, exclude hosts from it, or turn it off.
• Settings › Connection can point at a CA bundle, for a cluster or an inspecting proxy whose certificate authority the system does not already trust. Those certificates are added to the ones already trusted, never put in their place.
• Settings holds the chart repositories and OCI registries Kubermeister reads charts from. Each one is read when it is added and again whenever you refresh it, its index is cached on disk, and any username and password go into this system's keychain rather than the settings file.

Changed

• The manifest editor has Review changes beside Save: it shows what the save will change — the lines it removes and the lines it adds, against the object as it stands — with Save on that view, so a replace that carries the whole object can be read before it is written rather than after. Save itself still writes in one press.
• A new version is now downloaded in the background and installed the next time the app quits, rather than waiting behind a button nobody had a reason to press. Settings › Updates still offers being asked before anything is downloaded, or never checking at all. An install set to never check keeps that; one set to be asked moves to the new default, because earlier versions saved being asked whether or not anyone had chosen it.

Fixed

• The window no longer sits on screen empty and white while Kubermeister closes; it goes when quitting is confirmed.
• A cluster call that runs past the read timeout is cancelled instead of left running. A screen waiting on a slow or unresponsive cluster no longer leaves a request open on every refresh, and the credential plugin a call started — an SSO login waiting on a browser, say — is ended with it rather than piling up one process per retry.
• A container's log reads as the container wrote it. Every line used to be prefixed with a level the app worked out from the text, which repeated the level most services already print, and invented one for the lines that print none.
• The Linux AppImage starts on Ubuntu 23.10 and newer. It was built on a runtime that loads libfuse2, which those releases no longer ship, so it failed at launch until that library was installed by hand.
• A Linux install from the .deb package is told when a new version is released. It still never installs one itself, since the package belongs to the system, but it names the version and links to the page to get it from.

Fixed

• On Linux the window, the dock and the application switcher show Kubermeister's icon rather than the desktop's generic placeholder.

Added

• The Logs tab opens already following, on the newest lines, rather than waiting for Live.
• The log console's View menu holds how lines are read: whether they wrap, whether each one carries its timestamp, and how many lines of tail to read.
• The log search can highlight its matches where they sit and keep every line, for the lines around a match that explain it, instead of hiding every line it misses.

Fixed

• The log console follows the newest line as it arrives and hands the view back the moment you scroll up to read; it used to stay where it opened while lines piled up below.
• A workload's Logs tab shows each pod's whole name, where the column cut off exactly the suffix that tells one pod from another.

Fixed

• The cluster summary's events and alerts share the space below the health chart and scroll on their own, so the newest events stay in view without scrolling the page.

Added

• A Secret's Keys tab reveals one value when asked and masks it again after 30 seconds, and copies a value to the clipboard without ever showing it.

Changed

• The update popover, the toast and Settings no longer print a release's notes, which are a long list of pull requests; they link to the release page, and the native update dialog offers a Release Notes button.

Added

• How often Kubermeister checks for updates is a setting: every hour, every 4 hours (the default), every 12 hours or once a day.

Changed

• A kubeconfig that will not load no longer stops the app at an error card: the window opens as usual and the top bar says what is wrong, with ways to try again, choose another kubeconfig or go back to the default.

Fixed

• Check for Updates… in the menu works even when the window has not loaded, answering in native dialogs and downloading and installing from there.
• A context whose cluster or user entry is missing from the kubeconfig is named as the problem, and marked in the context selector, where every screen said "Something went wrong".

Fixed

• A kubeconfig with an entry that names no cluster, no user or no name at all loads the way kubectl loads it, with that entry left out, instead of failing as a whole and taking every working context with it.
• A credential plugin named by bare command, as aws eks update-kubeconfig writes it, is found when Kubermeister is started from the Finder or the Dock, and a plugin that is missing or fails says so by name instead of "Something went wrong".

Fixed

• The update popover shows a release's notes as text rather than raw HTML.

Added

• How long a cluster read may take is a setting, 60 seconds by default where it was a fixed 15, for clusters that need longer to list.

Removed

• The Namespaces and Nodes lists no longer show pod counts, and the cluster summary no longer totals pods. Each count listed every pod in the cluster on every refresh, which made large clusters slow and look unreachable.

Fixed

• A cluster that cannot be reached reads as unreachable rather than "Something went wrong", and one that is only slow says it timed out and points at the setting.
• A list that times out under All namespaces suggests choosing one namespace.

Changed

• A new icon: a ship's wheel, with its king spoke marked.

Removed

• The disabled Deploy button on the cluster summary.

Fixed

• Changing namespace closes the detail page of an object in the namespace just left, as switching context already did.
• The namespace selector shows that it is loading instead of offering All namespaces alone.
• Lists empty and reload the moment the namespace switches, rather than showing the previous namespace's rows for seconds.

Added

• A new version is announced in a toast and a badge in the top bar, with one click to download it, restart into it, or try again, and a setting to say whether it downloads by itself.
• Deployments, StatefulSets and DaemonSets restart from their detail screen.
• A Deployment rolls back to an earlier revision, pauses and resumes a rollout, shows one in progress, and compares two revisions side by side.
• Nodes cordon, uncordon and drain. The drain dialog shows what it will evict first, and the drain respects PodDisruptionBudgets and reports each pod as it goes.
• A Helm release rolls back or uninstalls, and stays a release the Helm CLI can read.
• A Job runs again, a CronJob runs now or has its schedule suspended and resumed, and a pod can be evicted or deleted without waiting.
• A pod links to the workload that runs it, and a workload lists the pods it owns.
• A pod's containers are listed with their part, init, app or ephemeral, and each one's usage beside what it requested.
• Describe gives pods and nodes the reading kubectl describe does, with copy and download.
• The log console searches by text or by pattern, with case sensitivity, reads a chosen container and time window, and keeps up to 50,000 lines, set in Settings, while drawing only those in view.
• A workload's Logs tab follows all of its pods at once, each coloured by pod.
• Port forwards keep running when you leave the page that started them, are listed and stopped from the top bar, can target a Service and survive its rollout, and are offered again per context.
• ReplicaSets, ReplicationControllers, PodDisruptionBudgets, PriorityClasses, Leases, RuntimeClasses, IngressClasses, the CSI kinds, webhook configurations, ValidatingAdmissionPolicies, APIServices and FlowSchemas have lists and detail screens.
• The instances of any custom resource definition are listed, with the columns the definition declares.
• A namespace has a screen of its own: what lives in it, its quotas and limits, and what its pods use against what they asked for. The Pods list groups by node or by workload.
• Every detail screen names the controlling owner and any finalizers holding a deletion open.
• Related shows what else an object is tied to, and why.
• Each list remembers which columns you chose.

Changed

• Screens watching the same list share one watch, and lists draw only the rows in view, so large lists stay fast.

Fixed

• Switching context ends every port forward, shell and log follow, which could otherwise carry on into the new cluster, and a write refuses to act on anything but the cluster and namespace on screen.

Added

• The first release that works with a cluster. Kubermeister reads your kubeconfig without changing it, and switches context and namespace from the top bar or the command palette (⌘K or Ctrl+K).
• A cluster dashboard with CPU and memory charts from metrics-server, alerts and recent events.
• Lists and detail screens, kept live, for nodes, namespaces, events, quotas and limit ranges; pods, Deployments, StatefulSets, DaemonSets, Jobs, CronJobs and HorizontalPodAutoscalers; Services, Ingresses, Endpoints and NetworkPolicies; PersistentVolumes, PersistentVolumeClaims, StorageClasses and VolumeSnapshots; ConfigMaps, and Secrets with their values masked; ServiceAccounts, Roles, ClusterRoles and their bindings; CustomResourceDefinitions and Helm releases.
• A pod's logs followed live, a shell into its containers, and port forwards to it.
• CPU and memory usage on the pod and node lists.
• A Manifest tab on every detail screen shows the object as the cluster holds it, with download.
• Objects can be created, edited, deleted and scaled.
• Settings, and light and dark themes.
• The window reopens where it was closed.

Added

• The first release: the application window, packaged for macOS, Windows and Linux, updating itself in the background and installing a new version when it quits.