The problem

Research conducted 2026-07-22 against GitHub issue trackers, Hacker News threads, and the state of the ecosystem. Reaction counts and issue ages are from that date.

The landscape shifted in January 2026

The official Kubernetes Dashboard was archived on 2026-01-21 and moved to the kubernetes-retired organization. SIG UI now points people to Headlamp as the successor. Headlamp therefore inherits the whole community's expectations, and its backlog reflects the load: 834 open issues, against 190 for Freelens and 135 for k9s.

What people complain about

Metrics are wrong or absent

The loudest category by a wide margin. Freelens's four most-upvoted open issues are all metrics:

Then the bug pile: #964 (pod shows double memory), #1111, #1555, #1883 (CPU renders as 0.0Ki). Headlamp carries the same class of defect in #2043. Lens had #7299 and #8011 years earlier.

Root cause: the entire Lens lineage hardcodes Prometheus query shapes. Point one of these tools at metrics-server or VictoriaMetrics and values double, zero out, or disappear.

Design consequence for kubeside: a metrics source is an interface with three implementations from day one, and metrics-server is the default.

Authentication eats the web UI

Headlamp's pain is concentrated here. Eight of its fifteen most-reacted open issues are auth:

• #1716 EKS SSO, open since 2024-02-14, 35 comments.
• #2848 OIDC redirect loop on every click with Azure AD.
• #4198 OIDC impersonation broken on EKS.
• #1582 Flatpak sandbox cannot execute aws eks get-token.

Design consequence: in local mode the binary runs kubeconfig exec credential plugins as a native child process with the user's real environment. The entire sandboxing class of bug disappears. In-cluster mode inherits the OIDC problem and ships later, once local mode is solid.

Logs across more than one pod

Freelens #687 and #1460 both request deployment, daemonset, and statefulset log views. k9s #1399 ("Stream closed EOF") has been open since 2021-12-23 with 34 comments. On Hacker News a user described opening six sessions manually to read one deployment's logs, and Aptakube's founder led his pitch with exactly this feature. Everyone else still shells out to stern.

Design consequence: whole-workload log tailing is the default. Per-pod is a filter applied to it, never the starting point.

Single-maintainer risk on k9s

34k stars. Contributor totals: derailed at 1121 commits, next human at 40. Of the last 100 commits, 43 are dependabot and 4 are the maintainer. Meanwhile sorting pods by CPU or memory has been broken since January 2026 across 28 comments, and decoding a secret without base64 gymnastics has been the top request since 2021-01-21.

Design requests exist, phrased as bug reports

Nobody files an issue saying "make this beautiful". They file legibility bugs and personalization gaps, because those are the available words:

• Freelens #1280 custom themes: 42 comments, funded with a 50 dollar bounty, and the entire thread is an assignment dispute between two contributors. Nothing shipped.
• Freelens #550 accent color, #186 filter box unreadable in the light theme, #469 font size.
• Headlamp #1788 "design papercuts umbrella issue", #3574 nodes table barely visible in dark theme, #1835 widescreen dashboard brainstorm where seven people typed /assign and nothing shipped.
• The largest thread on Headlamp's 2020 HN launch is a pile-on against hiding buttons based on RBAC instead of disabling them with an explanation. Six years later, unchanged.

Neither project has design capacity. Freelens tried to buy some for 50 dollars. Aptakube, K8Studio, and Kunobi charge money and compete purely on polish, which demonstrates willingness to pay for what the open source options do not deliver.

Design consequence for kubeside: controls a user lacks permission for render disabled with the missing RBAC verb named in a tooltip. Never hidden.

The four uncovered gaps

These define the product.

1. Resource-centric navigation

Every tool mirrors the API tree: Workloads, then Pods, then ConfigMaps. A developer thinks "the checkout service", spanning a Deployment, a Service, an Ingress, two ConfigMaps, a Secret, and an HPA. Kubevious named this years ago and stayed niche.

2. No sense of time

Everything renders now. "Why did my pod restart twenty minutes ago" has no answer in any of these tools. k9s #3599 asks for a diff against the previous version of an object. Google shipped KHI as a separate timeline viewer because no dashboard does history. Komodor sells this commercially.

The assumption everyone makes is that history requires a database, which is why no free tool has one and why Komodor charges for it. The assumption is wrong. Kubernetes already retains ReplicaSets, ControllerRevisions, Helm release secrets, pod termination states, and recent events. The history is sitting in the cluster and no tool assembles it. That is the real gap, and closing it needs no storage at all. See 05-architecture.md.

3. Effective configuration

Headlamp #4798 requests resolved environment variables for a running pod. No tool merges env, envFrom, ConfigMap, Secret, and downward API into one view with provenance per row. This is the most common developer question and every tool makes you assemble the answer by hand from four screens.

4. Operator sidebar, developer user

Nodes, PersistentVolumes, CRDs, Helm, and RBAC dominate the navigation. Headlamp #4051 ("improve UX for non-cluster-admins") acknowledges the mismatch. A developer with namespace-scoped RBAC sees a sidebar mostly full of forbidden links.

Validation

The grouping engine was tested against a real, organically grown cluster on 2026-07-23 rather than a fixture. 141 apps, of which 134 resolved on the first rule of the precedence chain. Nothing was split across rows and nothing unrelated was merged; the fallbacks were all vendor components that genuinely ship without recommended labels.

That settles the riskiest question in the plan: developers do recognise the output as their own services. It also surfaced two findings the plan had not anticipated, recorded as issues #39 and #40.

Competition

skyhook-io/radar launched 2026-01-20 under Apache-2.0 and reached 2.6k stars in six months, positioned as "the missing open source Kubernetes UI: topology, event timeline, and service traffic". Its comparison page attacks Lens, Freelens, k9s, and Headlamp on gaps 1 and 2.

Assume Radar gets topology right. The kubeside wedge is narrower and different: the developer's four questions, resolved configuration, whole-workload logs, and the same application compared across environments.

Reference points