Changelog

Aksimuth 0.17 finds EKS and GKE clusters beside AKS, gives cert-manager, Crossplane, Kyverno, Kueue and Fleet Manager views of their own, lists several namespaces at once, and shows more of what an object refers to without leaving it. Every change still goes through the same permission check, server dry run and one-use confirmation.

EKS and GKE.

• AWS and Google Cloud tabs in the fleet list EKS and GKE clusters through your own AWS CLI and gcloud: version, status, node groups or node pools, Auto Mode or Autopilot, and where each version stands in its support window.
• Neither cloud is asked anything until you pick a profile or configuration. Connecting builds the kubeconfig in memory; nothing is written to ~/.kube.
• When a sign-in has expired, Aksimuth names the command to run, such as aws sso login.

Platform dashboards.

• A Platform group appears on clusters that run cert-manager, Crossplane or Kyverno.
• Certificates with an expiry radar, the failing issuance chain down to the ACME Challenge and its reason, and the Ingresses and Gateways that use each certificate. Renew goes through the same preview as every other change.
• Crossplane claims, composite and managed resources as trees, with the failing resource and its message marked.
• Kyverno policies by Enforce or Audit, with their results and exceptions, and Gatekeeper’s constraints beside them.

Batch and fleets.

• Kueue queues with quota per flavor against what is reserved, and why each workload is not admitted. Hold, Hold and Drain, and Resume go through the preview.
• JobSets with their restarts, Jobs and Pods.
• On an Azure Kubernetes Fleet Manager hub: member clusters with their health, placements with each member’s progress and failed resources, and staged update runs.
• AKS Automatic clusters say so on the overview, with the guardrails that come with them.

Several namespaces at once.

• Choose up to 20 namespaces. Lists, live updates, Helm releases, status filters, export and selection follow them, and saved views keep them.
• Favourite namespaces stay at the top, and recent choices are kept. If you may read only some namespaces, each is listed on its own and the others are named.

More in place.

• A ConfigMap value shows beside the variable or volume that uses it, and an envFrom or a volume expands into its keys.
• Secret keys stay masked. Turn on Reveal Secret values, which asks you to confirm, and the eye beside a key reads that one value from the cluster, as you, and shows it for up to a minute. It is never stored, and at most 30 values are read a minute.
• A pod’s node and controller, a claim’s volume, the ready endpoints behind an Ingress or route, a workload’s autoscaler, image digests and each probe’s latest failure open where you are.

The agent in the window.

• With Agents may move the window turned on for a cluster, an agent can bring a view or an object to the front, never over an open sheet or in front of other apps, with a banner naming it.
• An agent can propose an extension or a saved view. It is checked, waits for your review, and is installed or saved only when you choose. Every proposal and outcome is in Agent Activity.

Managed by your organisation.

• A policy deployed by Intune, Jamf or Group Policy can mark clusters Production or Read-only, limit agent access, keep network features off, pin updates, and limit bulk changes, extensions and assistant providers. Settings shows what it sets. No server, no account.
• Every committed change is written to a local journal, never with a value, that you can export and your organisation can have copied to a folder it names.

Smaller things.

• A sidebar group closes even while one of its lists is showing.
• File transfer progress moves smoothly.
• An extension’s manifest error names the exact entry it is about.
• The window runs under a strict Content Security Policy.

Aksimuth 0.16 works on many objects at a time, remembers the objects and lists you come back to, signs in to clusters whose credential plugins ask for a code, and reaches further into AKS: GPUs, Log Analytics and Linux on Arm. Every change still goes through the same permission check, server dry run and one-use confirmation.

Change many at once.

• Select rows in any list and restart, scale, cordon, suspend, label or delete them together. Each object is permission-checked and dry-run first; one preview shows what each will do and which ones won’t, and you confirm once for exactly that list.
• A delete is bound to the very object you previewed, so a pod re-created under the same name is never removed by mistake. Deleting namespaces, nodes or volumes asks for their names.
• Force Delete for a stuck pod explains what it does and never removes finalizers.

Keep what you watch.

• Pin objects and kinds to the sidebar with their live status.
• Save a filtered list (namespace, selector, filter, sort, columns, status) under a name, for one cluster or any.
• Filter pods by status (Running, CrashLoopBackOff, ImagePullBackOff and more) and workloads by health, and choose default columns for every cluster.
• Split View puts two clusters, or two views of one, side by side, each with its own Production and Read-only state.

Sign in where you work.

• A kubelogin device code is shown in the window, with Copy Code and Open in Browser, and an exec plugin’s password or MFA prompt is answered there too. What you type goes only to the plugin and is never stored.
• A plugin that would wait unseen is stopped and the cluster asks you to sign in again, instead of hanging.
• Azure Government and Azure China clusters are discovered like any other.

Further into AKS.

• Cluster › Accelerators: which nodes have GPUs and other accelerators, who holds them, which pods are waiting for one and why, driver and device plugin health, and DCGM metrics where they exist.
• Log Analytics, off until you turn it on: logs and events from Container Insights for pods that are gone or ranges the node no longer keeps, a metrics fallback, and a read-only KQL box. Agents can read the same records by choosing a table, filters and columns; Aksimuth writes the query.
• Linux on Arm64, with updates.

Lighter.

• Large lists scroll more smoothly and use less memory while they do, open lists read their objects once, and an idle window does almost nothing.

Aksimuth 0.15 looks across clusters as easily as into one, opens any view from a link, checks how workloads sign in to Azure, shows traffic down to the request, and lets you teach it about the custom resources you run, without running anyone else’s code. Everything new reads; nothing new changes a cluster.

Fleet.

• Fleet lists one kind, built-in or custom, across every connected cluster in one table, with each cluster’s colour and tag. A cluster that refuses, is unreachable or doesn’t serve the kind says so in its own row; the others still answer.
• Compare any two objects of a kind, in one cluster or two: a clean YAML diff that leaves out what each cluster assigns on its own, and a summary of replicas, images, requests, limits and environment variable names for workloads. Secret values never reach the window; only whether each key is the same.

Links.

• aksimuth:// links open a cluster’s view or a single object on a chosen tab. Copy Link is in the inspector and the toolbar.
• A link only navigates. It never connects a cluster without asking, never starts a terminal or a port forward, and never changes anything.

AKS › Identity.

• Checks Entra ID integration, Azure RBAC, local accounts, the OIDC issuer and Workload Identity for the cluster, and how your kubeconfig signs in.
• Follows Workload Identity from ServiceAccount to pod: missing annotations, pods the webhook did not inject, and leftover pod-identity objects, each with what to do.
• On request, checks each identity’s federated credentials against the cluster’s issuer and ServiceAccounts, read-only through your Azure CLI.

Traffic and metrics.

• Traffic › Requests shows HTTP, DNS and Kafka flows from Hubble where L7 visibility is on: request and error rates, p50 and p95 latency per workload, and DNS failures. Query strings, headers and bodies are never kept.
• The Metrics tab and the cluster overview can read from an in-cluster Prometheus or an Azure Monitor workspace, with history up to 24 hours and a read-only PromQL box. Azure Monitor is off until you turn it on in Settings.
• Agents get two new read-only tools: the identity audit and metric queries, which stay confined to the cluster they were enabled for.

Extensions.

• An extension is a single reviewed file that adds columns, Summary sections, health, links to related objects, https links and shortcuts for the kinds you choose. Install it from a folder, read what it asks for, and turn it on per cluster.
• Extensions run no code, make no requests and cannot change a cluster. What they show stays in the window; agents and the assistant never see it.

Faster to start.

• Your kubeconfig clusters appear without waiting for the Azure CLI, in about a second on a current Mac; the Azure list follows when the CLI answers. Live tables also redraw less on every change.

Aksimuth 0.14 makes each cluster harder to change by mistake, lets the agents you already run read your clusters through Aksimuth on your terms, and covers more of what a cluster runs: Flux and Argo CD, Gateway API, devices, admission policies and AKS upgrades. Everything new that changes a cluster goes through the same permission check, server dry run and one-use confirmation as before.

Cluster guardrails.

• Each cluster can carry a colour and a short tag, shown on its rail button, along the top edge of the window and in the window title.
• Mark a cluster Read-only and Aksimuth itself refuses every change to it. Type its name to unlock it for 5 to 60 minutes; it locks again on its own.
• Only clusters marked Connect at launch connect when Aksimuth starts.
• KUBECONFIG is read from your login shell when Aksimuth is opened from the Dock or a launcher, and Settings shows where every kubeconfig file came from.
• Update checks have three modes: Off, Notify, and Automatic download.

Agent access.

• Aksimuth can serve your clusters to the MCP clients you already run, such as Claude Code, Copilot CLI or Cursor, over a private socket or the loopback address, with a challenge before the token is sent. There is no account and no relay.
• Each cluster is off until you enable it, and read-only unless you allow changes. A change an agent proposes waits for your approval in the Aksimuth window; on a Production cluster, for its name typed. Changes to Secrets are refused unless you allow them.
• Every call is kept in a local ledger under Observe › Agent Activity, and can be exported.

Pods, files and debugging.

• A pod’s Files tab browses a container, downloads files and folders, and uploads with a preview of what it writes and replaces.
• Debug adds an ephemeral container, copies a pod, or starts a debug pod on a node, with kubectl’s profiles, then attaches a terminal.
• Resize changes a pod’s CPU and memory in place.

More of the cluster.

• Flux and Argo CD get their own panels and an overview. Objects they manage carry a Managed by badge and a warning before a manual edit, and Reconcile, Sync, Suspend and Resume are actions.
• Gateway API, dynamic resource allocation devices, and validating and mutating admission policies are first class.
• AKS › Upgrade Radar places version support, upgrade paths, node image retirements and deprecated APIs for one cluster, or for the whole fleet. Node Auto-Provisioning (Karpenter) and Deployment Safeguards have views of their own.
• A Security tab reads Trivy and policy reports where the cluster has them, and checks Pod Security and who can do what on this computer. A Cost tab compares requests with usage, uses Azure list prices if you turn them on, and shows OpenCost where it runs.

Logs and YAML.

• One log viewer for a pod, a workload or a label selector. It follows rollouts, reads JSON and logfmt as fields, filters by level, text or regular expression, draws a volume histogram you can narrow by time, links trace IDs, and saves what you see.
• A schema-aware YAML editor with completion and checks from the cluster’s own OpenAPI schema, diffs against the live object and the dry-run result, and a rebase when the object changed underneath you.
• Edit YAML now replaces the object by default, as kubectl edit does. Server-side apply of only the fields you changed is an option. An edit that still holds a masked value is refused.

An assistant with your own model.

• Besides Claude Code and GitHub Copilot CLI, the assistant can use your own Anthropic, OpenAI-compatible (including Azure OpenAI) or Ollama endpoint. Keys are kept in the operating system’s keychain.
• Explain This starts from a failing pod, a warning event or a failed GitOps object. Conversations are kept on this computer, redacted, and export as Markdown.
• Redaction is tighter: annotations that carry a copy of an object, Helm values in Flux and Argo CD objects, and error messages are redacted before anything is sent, and an exported conversation opens without fetching anything.

A namespace at a glance. Open a namespace and it starts on an Overview: workloads by kind with their health, pods that are not ready, counts for services, ingresses, routes, config, volume claims and policies, quota use, limit ranges, warning events, Helm releases, and the busiest pods. Every tile and row opens the object, or the list scoped to that namespace.

Relationships in both directions. Each object’s summary groups what it owns, uses, routes, selects, grants and runs on, and now what uses it: a ConfigMap, Secret, volume claim, service account, priority class, storage class, Role or Service names the pods, workloads, bindings and routes that depend on it. A Related tab draws the tree two or three steps out, with status on every node, and you can walk it from the keyboard.

Details for every common kind. Pods show IPs, QoS, containers, probes, resources, env sources and volumes; Services their ports and endpoints; Ingresses their rules; ConfigMaps their data; Secrets their key names and TLS certificate details, never a value; Nodes, storage, HPAs, Jobs and CronJobs their own. Click a label or a selector to narrow the list to it, shown as tokens you can remove.

Helm charts, with the same care as everything else. Add HTTPS chart repositories, or import Helm’s own list read only, search their charts, and install one or move a release to another chart version. Before you confirm, Aksimuth checks your permission for every kind the chart touches and sends every rendered object, hook and CRD to the API server as a dry run, as you, then shows the outcome per object. OCI references work the same way. Repositories are kept in Aksimuth’s own settings; ones that need a password are not supported, because Aksimuth stores no credentials.

Lists that keep up.

• The command palette searches every kind you may list, by name, grouped by kind.
• Load More continues past the first 500 rows.
• Pods gain Controlled By and CPU and memory columns; column choices are kept per cluster.
• Export as CSV writes the rows shown. Save Log writes a container’s or a workload’s whole log.
• Create Resource starts from a template or your YAML; every document is permission-checked and dry-run on the server before a one-use confirmation.
• A dot beside each cluster shows whether its API server is answering.

Port forwards you can keep. Choose Save on a running forward and Aksimuth keeps it: the cluster, namespace, pod or service, port and a preferred local port, never a credential. The running forward carries on at the same address. Start a saved forward again from Port Forwarding, the object’s Forward tab or the command palette. If its preferred port is busy, it takes a free one and says so. Turn on Restart on reconnect and it starts again after a pod restart or a lost connection. Saved forwards are kept on this computer.

Open in Browser. A running forward to a web port (one named http, https or web, or port 80, 443, 8080 or 3000) offers Open in Browser. It opens only the forward’s address on this computer.

Hygiene findings, read only. Observe › Metrics › Hygiene lists what is worth fixing in the workloads in scope: missing requests or limits, :latest or untagged images, privileged containers, hostNetwork and hostPath, runAsNonRoot left unset, missing readiness probes, single-replica Deployments without a PodDisruptionBudget, Services that select no pods, and crash loops. Each finding has a severity and a reason, and opens its object in the inspector. Nothing on the tab changes the cluster.

Smaller things.

• Refresh Sessions keeps keyboard focus while the sessions reload.
• The keyboard focus ring stays visible with forced colours (high contrast) on.

Quick fixes. When Mermaid stops at an error, Cascadia offers a fix for the common mistakes: a misspelled or missing diagram type, a mistyped statement keyword where the type should be (clas Order, stat Idle), a one-dash flowchart arrow, a block without its end, a node named end, a label or subgraph title that needs quotes, a stray quote in an unquoted link label, a comment after code on a flowchart line, and an unclosed quote or bracket. Press ⌘. (Ctrl+. on Windows and Linux), or choose the fix from the error’s hover or the lightbulb. A fix is applied only when you choose it, and one Undo takes it back. Fixes are worked out in the browser; your code never leaves it. The full list is in the documentation.

New from Template. A gallery with a starter diagram for each of the 23 diagram types Cascadia supports, from the New and More menus and from Quick Insert. A template fills an empty diagram or opens a new one; it never replaces code that is already there.

Clearer colours. Pie slices, XY chart series, Venn sets, radar curves, mindmap branches, timeline periods, kanban columns, treemap cells and Git Graph branches take a palette that reads clearly in both themes.

Smaller things.

• A new diagram’s first save keeps the editor’s undo history.
• At phone width, a diagram created from the Files drawer closes the drawer and returns focus to the Files button.

The editor in this release is Cascadia UI 1.2.0.

HSA Ledger, which is in development, now shows each corporation’s activity. The Activity page lists what was recorded, newest first, in plain language: who did it, what it was about, when, and the outcome. Show earlier brings in older events. The page is read only, and it is offered only to roles that may read the corporation’s audit log.

In the RemuNation preview, approving, finalizing and voiding pay runs works again. Each of these actions had been refused with an error; they now go through as expected.

Go anywhere by name. ⌘K (Ctrl+K on Windows and Linux, or ⌘⇧P / Ctrl+Shift+P) opens a palette that finds views, kinds, namespaces, clusters and resources. It only navigates; it never changes anything.

Tables you can drive from the keyboard. Arrow keys move through rows, Enter opens the inspector, / jumps to the filter, and ⌘[ and ⌘] (Alt+← and Alt+→ on Windows and Linux) move back and forward through what you opened.

Production guard rails. Give any cluster an alias and a colour, and mark it as Production. Every change to a Production cluster (scale, restart, rollback, apply, delete, suspend, Helm) then asks you to type its name first, and that check is enforced by the app itself, not only the screen. Labels are kept on this computer; your kubeconfig is not changed.

More of your clusters.

• Add kubeconfig files and folders in Settings, read only.
• Logs, exec and attach follow the kubectl.kubernetes.io/default-container annotation.
• Custom resource tables show the columns their definitions declare.

Calmer, clearer motion.

• Dialogs, sheets and the inspector ease in and out.
• Lists keep their rows while they refresh, and loaders only appear when a load is slow.
• Rows a live watch adds or changes are highlighted briefly.
• With Reduce Motion on, movement becomes a short crossfade.

An editor that knows Mermaid. Syntax highlighting, completion for diagram types and keywords, and an error marker on the line that stops a diagram from rendering.

Exports that match what you see. PNG exports keep multi-line labels and dark diagrams’ colours, at twice the resolution, on the theme’s background, white, or transparent. You can also copy a diagram as PNG or as SVG markup straight to the clipboard.

History you can see. Version history shows the code change and the rendered diagram before and after.

Smoother canvas. Switching diagrams or themes crossfades instead of flashing, and zoom and fit commands ease into place. With Reduce Motion on, they change instantly.

Follow any reference. Owners, relationships, an event’s object, a pod’s node, Helm resources, traffic endpoints, segmentation findings and the pods behind a workload’s metrics and logs are all links. A link takes you to the object’s own list with it selected, and Back returns to where you were, on the same tab. Namespaces in any list narrow the scope to that namespace, and the overview tiles open their lists.

More relationships.

• HTTP and gRPC routes show the Gateways they attach to and the backends they route to.
• Gateways and Ingresses show their class; pods show their priority and runtime class.
• EndpointSlices show their Service and the pods behind them.
• NetworkPolicies show the pods they apply to.
• Deployments and CronJobs show the ReplicaSets and Jobs they own, and pods show the Services that select them.

Safer updates. Update manifests now expire and can be signed by a rotated key, and the installer is checked again right before it runs. A failed update on macOS or Linux reopens the version you had.

Fixes.

• Edit YAML refuses to save over a change made after it loaded, and never writes masked Secret values back.
• Contexts that sign in with a device code use your Azure CLI sign-in instead.
• The Assistant masks Secret and ConfigMap data and connection strings before a resource is sent.
• Switching the inspector from one object to another no longer fails.
• ⌘H hides Aksimuth, and quitting cleans up terminal kubeconfig copies.

Copy and paste in terminals. Pod shells and the cluster terminal now copy and paste the way your platform’s terminal does:

• macOS: ⌘C and ⌘V.
• Windows: Ctrl+C with text selected, Ctrl+V, and the Shift and Insert variants.
• Linux: Ctrl+Shift+C and Ctrl+Shift+V.

Right-click for Copy, Paste, Select All and Clear, and an optional copy-on-select. Pasting several lines asks first and shows a preview.

Export diagrams. The Cluster Map, Application Map and Traffic view export as Mermaid, to copy or save as Markdown. Files open directly in Cascadia. Cluster exports show node pools, nodes, workloads and zones, with shared nodes highlighted.

Tidier navigation.

• Service mesh detection and Kiali moved into the Traffic view.
• The Metrics view no longer repeats the overview’s per-node table.
• The second “Overview” in the sidebar is now “Workload Health”.
• Edit YAML focuses on editing; scale, restart and delete stay in the action bar.

After Restart to Update on Windows, Aksimuth could reopen with its window hidden, so it looked as if it had not restarted. The updater now reopens Aksimuth the way a double-click does. The app also makes sure its window is visible when it starts, so the update to 0.8.2 reopens it too.

While the new version installs, a small “Installing Aksimuth” window now shows progress, and it closes when Aksimuth reopens. It appears from the update after 0.8.2 onwards, because each update is run by the version being replaced.

Aksimuth 0.4 is a redesign against Apple’s Human Interface Guidelines: the toolbar sits in the window frame with a cluster › namespace scope path, the sidebar is a proper source list over the window’s material, there is a real menu bar with the shortcuts you expect, tables sort and alternate, and the app follows the system’s light or dark appearance instead of forcing its own. It can also keep itself current: turn on automatic updates once and it fetches a signed release manifest daily, downloads the new version in the background, and asks before restarting. Details of exactly what is sent are in Settings. The public build now carries a single, labelled support card at the foot of the sidebar, a link to contribute if Aksimuth saves you time; the ad-free build has none.

A small release for people on Windows: helper programs Aksimuth runs, including kubelogin when a cluster token refreshes, no longer flash a console window. They now start hidden, and credential plugins go through Aksimuth’s own quiet proxy. The Linux archive gains a launcher entry and an installer script so the app opens from the application menu rather than a terminal. Installed copies of 0.4 that turned on automatic updates will pick this up on their own.

Restarting to install an update on Windows no longer opens console windows: the installer step now runs inside Aksimuth itself. The inspector and sidebar can be resized by dragging their edge (or with the arrow keys), and the inspector can be widened for long logs in one click. Sidebar groups collapse. On Windows and Linux the menu strip is gone in favour of a Settings button and the usual Ctrl shortcuts; macOS keeps its menu bar. Node pools from GKE and EKS are now labelled alongside AKS.

Aksimuth 0.5.1 adds the three things people asked for most. The Logs tab of a deployment (or stateful set, daemon set, job or service) now follows every pod it owns in one view, tagged by pod, the way kubectl logs deployment/x --all-pods -f does, and keeps following through rollouts. The new Application Map draws each namespace as a tree, from ingresses and routes through services to workloads and pods, with health on every card. Services can be port-forwarded directly, and when Istio and Kiali are installed, Kiali opens in your browser in one click. On Windows, kubeconfigs inside WSL distributions are now found automatically, and the sidebar scrolls properly on small windows. Followed logs now appear line by line as they are written, where quiet pods previously showed nothing until enough output had built up.

On Windows, Restart to Update now shows the standard permission prompt before installing, as any installer does, instead of failing quietly and reopening the old version. If an update ever does not install, Aksimuth says so and offers to try again. Copies of 0.4.3 to 0.5.1 on Windows cannot install this update themselves: download 0.5.2 from this site once, and updates work from then on.

Installing over a running copy on Windows stopped with “Error opening file for writing” and an Abort, Retry, Ignore prompt, because Windows cannot replace a program that is still open. The installer and uninstaller now close Aksimuth before copying files.

On Windows, Restart to Update closed Aksimuth and then appeared to do nothing. The request for administrator rights came from a background helper, and Windows shows such a request only as a flashing taskbar button. Aksimuth now asks for permission itself, while its window is in front. If you decline, Aksimuth stays open and you can try again.

The dividers between the sidebar, the content and the inspector now show a small grip, so it is clear they can be dragged to resize.

Copies from 0.4.3 to 0.5.3 need one manual install from this site. Later updates install from the app.

When your Kubernetes identity is allowed to change a resource, Aksimuth now shows the controls for it under the resource’s title and in a context menu on every table row:

• Deployments and StatefulSets: Scale, Restart, Stop and Start. Stop scales to zero and Start brings back the size it had.
• Deployments: Roll Back to an earlier revision, with its images and date.
• CronJobs: Run Now, Suspend and Resume. Jobs: Suspend and Resume.
• Nodes: Cordon, Uncordon and Drain. Drain respects pod disruption budgets and leaves DaemonSet pods in place.
• Pods: Restart, for pods a controller manages. Every resource: Edit YAML and Delete.

Aksimuth checks each permission with the cluster before showing a control, and names any that are missing. Every change is validated by the API server with a dry run before you confirm it.

Tables now update on their own every few seconds while you look at them. The button beside Refresh pauses this. Every kind of resource in the sidebar now has its own icon.

Aksimuth can also send an anonymous daily count of copies in use. This is off unless you turn it on. See the privacy page for exactly what is sent.

Cluster Map. A new view of the cluster as it runs on machines: node pools, their nodes, and the pods on each node. Pods can be coloured by security zone, namespace, workload or health. Nodes can be grouped by pool or by availability zone. Clicking a node or a pod opens it with its controls.

You define zones for each cluster by namespace name, namespace labels or pod labels, and mark the ones holding regulated data as sensitive. Aksimuth suggests zones from the cluster’s own labels and pool names. The segmentation review reports:

• nodes shared across zones;
• sensitive workloads whose placement is not enforced;
• pools without a fencing taint, and workloads that tolerate that taint;
• shared agents on sensitive nodes;
• sensitive namespaces without a NetworkPolicy;
• anything that exposes a sensitive zone outside the cluster.

Each finding comes with a suggested fix, and the review copies as a report. Use it for PCI DSS segmentation reviews and similar work. It is a working aid, not an attestation.

Terminal. A shell for each connected cluster, in which kubectl, helm and the cloud CLIs already target that cluster and namespace. It uses a private kubeconfig that is removed when the terminal closes.

Helm. Release details, values (hidden until revealed), manifest, notes and history, with Roll Back, Edit Values and Uninstall. Every change is previewed first.

Live tables and overview charts. Tables follow the cluster through Kubernetes watches instead of reloading. The overview charts CPU and memory usage against requests and capacity for the last hour, for the cluster and for each node. A node’s details now list its pods.

Pods, workloads and nodes have a new Metrics tab with CPU, memory and network charts. When the cluster runs Prometheus, Aksimuth finds it and shows up to 24 hours of history. It queries Prometheus through the Kubernetes API, so no port forward is needed and your cluster permissions apply. Without Prometheus, the tab samples Metrics Server while it is open.

On Windows, Restart to Update installed the new version but did not reopen Aksimuth afterwards. When the installer closed Aksimuth, it also closed the helper that reopens it. The installer now closes only Aksimuth itself. The version shown under the cluster icons in the sidebar is now the real one.

Updating from 0.7.0 still closes Aksimuth without reopening it. Start it again from the Start menu. Updates after 0.7.2 reopen it by themselves.

The Cluster Map showed a blank window on clusters where Aksimuth could not suggest any zones. The Assistant did the same when a question could not start, for example when the local assistant was not signed in. Both now show their content, or the reason the question failed. If a view ever fails again, only that view shows an error, with a button to copy the details for a report.

The CPU and memory charts on the cluster overview drew samples older than an hour outside the chart. They also joined the times the overview was closed with a straight line, and clicking a chart left a black outline around it. The charts now show only the last hour, leave a gap where no samples were taken, and have no outline on click.

A new Traffic view, under Observe, shows which workloads call which, live.

Network flows stream from Hubble Relay on clusters with Cilium, including AKS with Advanced Container Networking Services, Retina and GKE Dataplane V2. You see every new connection, DNS lookup and HTTP request between workloads, and every call a NetworkPolicy dropped, with the reason. Calls appear on a graph from caller to callee, with thicker lines for busier paths and dropped calls in red. Selecting a workload or a line filters the live flow list below it.

Service mesh shows request rate, error rate and p95 latency between workloads, from the metrics Istio or Linkerd already record in Prometheus.

Aksimuth connects to Hubble through a port forward and only reads, so your cluster permissions apply.

Aksimuth 0.2 keeps every connected cluster in the left rail, so switching between AKS clusters is one click and never disconnects the others. Navigation now follows the shape people know from Lens: Jobs, CronJobs, DaemonSets, StatefulSets, ReplicaSets and the config, network, storage and access-control kinds are all first-class entries, and every list starts in all accessible namespaces. A new Nodes view, a Workloads overview and a cluster-wide Log Search (plain text or regex across every pod in scope) round it out. Downloads are on the product page; builds remain unsigned for now.

The local assistant in Aksimuth now works at cluster level and in both directions. Ask about the whole cluster or one resource; when the answer depends on something the assistant has not seen, such as the logs of a pod it found referenced in another pod’s logs, it asks Aksimuth for exactly that, and Aksimuth fetches it read-only, redacts it, and shows the request and the result in the transcript. You choose whether each batch waits for your approval or runs on its own within a fixed budget. Same rules as before: local CLIs only, no tools of their own, nothing leaves the machine unredacted.

RemuNation joins the catalog: Canadian payroll for small business — pay runs, CRA remittances, T4s, and direct deposit. It is a pre-release preview, not yet publicly available, and not certified by or affiliated with the CRA. Updates appear here in the changelog.

iscreambyte.com is live. It introduces two projects: Cascadia, a collaborative text-first diagramming app now in preview, and Aksimuth, a local-first desktop console for AKS and Kubernetes, with early builds in the works.