# KubeNx > KubeNx is a free, native Kubernetes desktop workspace for macOS, Linux and Windows, built in Rust with GPUI (the GPU-accelerated UI framework behind Zed). It brings Argo CD, Trivy security findings, Prometheus metrics, OpenCost allocation, logs, terminals, port-forwards, YAML and Helm into one app, with an optional AI assistant that works with your connected cluster. - Website: https://kubenx.app/ - Download: https://kubenx.app/download - Status: early development, free to use - Connects with your existing kubeconfig. No account and no server component. ## Install ### macOS (Apple Silicon and Intel) With Homebrew: ```sh brew install --cask kubenx/tap/kubenx ``` The app is unsigned and not notarized. The Homebrew cask signs it on your Mac during install, so it opens without being blocked. Or download the `.dmg` from https://kubenx.app/download, open it and drag KubeNx.app to Applications. Choose Apple Silicon for M-series Macs, or Intel. ### Linux (x86_64 and ARM64) Download the `.tar.gz`, extract it and run `kubenx`. Run `python3 install-desktop.py` from the extracted folder to add a menu and dock icon. Built for Ubuntu 24.04 compatible system libraries; a working Vulkan graphics driver is required. ### Windows (x86_64 and ARM64) Download the `.zip`, extract it, then launch KubeNx. Windows code signing is not configured. ### Release files The latest version and every download URL are listed in https://dl.kubenx.app/latest/manifest.json. Files are named `kubenx-{version}-{macos|linux|windows}-{arm64|x86_64}.{dmg|tar.gz|zip}` and are covered by a `SHA256SUMS` file in the same release folder. ## Features ### Argo CD GitOps, with the whole tree in view. See every application's sync and health at a glance, then follow one down to the pods it owns. Argo CD runs in your cluster; KubeNx reads it with your own kubeconfig. - A dashboard of sync and health totals, ApplicationSets, Projects and the applications that need attention. - A resource tree built from Kubernetes ownership: Application, Deployment, ReplicaSet, Pod, Service and EndpointSlice. Zoom, collapse branches, filter by kind, or search. - Refresh, or Sync selected applications after choosing exactly which resources to deploy. Pruning and force stay off. - Open the Argo CD UI through its ingress or a loopback port-forward. KubeNx never fetches admin secrets. ### Security Findings from your scanner, ready to triage. KubeNx reads the Trivy Operator reports already in your cluster and turns them into one searchable table. It never scans on its own and never treats a missing report as a clean bill of health. - Vulnerabilities by image, configuration audits by resource, and RBAC assessments by role, each with severity counts. - Filter by severity, namespace or scan age. Search by CVE, package or title. - Open a finding for its fixed version, advisory link and the owning workload. Owner details are checked against Kubernetes UIDs. - Export the filtered view to CSV. Coverage warnings show when a report family is denied or truncated. ### Monitoring Metrics where you work, with no port-forward. KubeNx finds Prometheus or VictoriaMetrics in the cluster and queries it through the Kubernetes API, so history, targets and alerts sit next to the workloads they describe. - CPU, memory, network and filesystem history for pods, workloads, nodes and the whole cluster. - Scrape targets, reachable or failing, plus a PromQL box with instant and range queries. - Bounded, read-only requests through the Service proxy. No arbitrary URLs and no credentials stored. - Optionally install a pinned kube-prometheus-stack after reviewing a dry-run of every object it would create. ### Cost Know what each namespace costs. Connect an OpenCost service and see allocation by namespace over 24 hours, 7 days or 30 days. Figures are labelled as estimates, not invoices. - Summary cards, a daily stacked chart by namespace, and CPU, memory, storage and network cost per namespace. - Idle cost on or off, sorting, and namespace filters. - Source totals are checked against their components, and stale data stays marked as stale. - Ask the AI assistant for savings ideas. Anything it proposes is a suggestion for you to review. ### Workloads The everyday work, and the broken pods first. Live tables for every resource kind, with the usage and status you actually look for. - Pods, Deployments, Services and custom resources, with filters, saved column layouts and CSV export. - Live CPU and memory beside limits, when metrics-server is available. - Container logs, pod and local terminals, and Pod or Service port-forwards. - Several clusters open at once, each keeping its own watches, tunnels and layout. ### Everyday workspace - Native speed: a GPU-accelerated interface powered by GPUI. - All your clusters together: switch workspaces without closing background watches, logs, terminals or port-forwards. Navigation and layouts are restored after a restart. - Resources without the digging: search and filter workloads, inspect YAML and resource relationships, and check node health, events, CPU and memory in one place. - Command palette, customizable table columns, CSV export and themes. - Helm and YAML: browse Helm releases, review YAML before you apply it, and keep events and node health one click away. ## AI assistant - Answers with live context: ask about pods, events, metrics and workload health. The assistant inspects your connected cluster and can open the relevant views for you. - Your AI, your choice: connect Claude Code or Codex with your existing subscription, or bring an API provider or a local model through an OpenAI-compatible endpoint. - You approve the changes: sensitive reads are reviewed before data reaches your provider, and every proposed workload restart or scale needs your confirmation, even with Auto mode on. - AI agents need Node.js 22+ (npx) or Bun installed. Setup: open Preferences → AI providers, connect Claude Code or Codex (or add an API provider or local model), then open the assistant and ask about your connected cluster.