SheltieStar on GitHub

UI/UX competitive analysis

How Sheltie's user experience compares with the tools homelab users already run, what we lack, and the design decisions that follow. This complements the feature comparison in COMPETITIVE_ANALYSIS.md. The design rules themselves live in DESIGN.md.

Observations about other projects are based on their public demos, documentation and community write-ups as of October 2026, and are kept deliberately general. UIs change often, so check each project for its current state.

The question each tool answers

Tool The question its home screen answers Typical first impression
Uptime Kuma 2.x "Which of my monitors are up?" Polished monitor list with heartbeat bars and a detail pane. Strong status pages, 90+ notification providers configured in the UI.
Beszel 0.19 "How are my servers doing?" Clean table of systems with resource meters, per-system history charts (1h to 1 week), ZFS pools, and an iOS companion app with home-screen widgets.
Maintenant "Is my whole Docker host healthy?" One binary: auto-discovered containers, endpoints, TLS certificates, image updates, exposed-port warnings and a status page.
Netdata "What is every metric doing right now?" Very rich, very dense chart wall aimed at experts.
Gatus "Are my endpoints healthy?" Minimal, read-only status list driven by YAML.
Dozzle "What is this container printing?" Live, searchable multi-container log viewer.
Dockge / Portainer "What is running, and how do I change it?" Compose stacks and containers with start/stop/update controls and logs.
Pulse "How is my Proxmox cluster?" Unified view of Proxmox VMs, backups, Docker and Kubernetes with threshold alerts.
Homepage / Homarr "Where are my services?" Start pages with tiles and widgets; Homarr leads on drag-and-drop approachability.
Sheltie "Is everything OK, what happened while I was away, and what should I do?" One sentence, the problems with buttons to fix them, then the story since your last visit.

Dimension by dimension

UX dimension Uptime Kuma Beszel Maintenant Dozzle / Dockge Sheltie today
3-second "am I OK?" Scan the list for red Scan the table Dashboard summary n/a Plain-language status sentence and glyph on every page
"What happened while I was away?" Per-monitor event history Limited Event list — "Since your last visit" summary on Overview
Next step for a problem Mostly informational Informational Informational Start/stop/restart, logs Every problem carries buttons (but no logs yet)
Diagnosis (logs) — — — ✅ core feature ❌ missing
Incident lifecycle Per-monitor timeline, maintenance Alert history Alerts — Flat event list, no grouping or acknowledge
Correlated root cause — — — — ❌ duplicates (site down + its container stopped)
History ranges 24h / 30d uptime 1h → 1w charts Yes — Last 80 checks; host trend only "since you opened"
Check types 20+ (HTTP, TCP, DNS, ping, push…) n/a HTTP, TCP, heartbeat, TLS n/a HTTP only
Notification setup 90+ providers in the UI with "Test" Webhooks/email in UI In UI — Telegram via config only
First-run onboarding Create admin, then empty list Copy-paste agent command Zero-config auto-discovery Zero-config Checklist, but the token comes from docker logs
Mobile Usable, desktop-first Responsive + iOS app Responsive Responsive Mobile-first: bottom tabs, sheets, 44px targets (no push yet)
Keyboard / power use Limited Some search Some Search ⌘K palette, g-shortcuts, ? help
Stale / offline honesty Reconnect indicator — — — Banner + last-known snapshot with its age
Destructive actions Confirms deletes — — Confirms Confirms every restart/remove/clear; Undo on remove
Accessibility Color-coded bars Partial Partial Partial WCAG 2.2 AA enforced by axe in CI; status never color-only
Visual identity Green accent, dark default Neutral component-library look Neutral dashboard Utility Ink & Cream: cream canvas, ink actions, color only for status, sheltie mark
Two-way control from phone Notifications only Notifications + app Notifications — Telegram commands plus a phone-ready dashboard

What we lack

Sheltie is a monitoring and rescue tool. The gaps below are ranked by how much they weaken the rescue loop: notice → understand → fix → confirm it's fixed. Each one is tracked in ROADMAP.md.

P0: gaps in the rescue loop (the core promise)

  1. Correlated problems. Today "Blog is down" and "blog has stopped" appear as two separate problems with two different buttons. They are one incident: "Blog is down because its container stopped", with a single Start action. No competitor does root-cause grouping for homelabs, so this becomes part of the moat. (src/lib/insights.ts)
  2. Technical copy leaks through. "502 Server Error: Bad Gateway for url: https://…" and key-value event details ("Site: Blog · URL: … · Expected: 200 · Status: 502") break our own voice rules. Rewrite as "Returned 502 Bad Gateway" and "Blog went down: it answered 502 instead of 200". (insights.ts, format.ts)
  3. No logs in the rescue flow. Dozzle, Portainer and Dockge all show logs. Without them, "Start" is a guess. Add View logs (last 200 lines, search, copy, follow) as a sheet on every container and on container problems. Needs a /api/docker/logs endpoint.
  4. Flat incident list. Group events into incidents with start, end, duration, what auto-heal tried, and an Acknowledge action that stops repeat notifications. Model: Kuma's per-monitor timeline plus PagerDuty-style acknowledgement.
  5. First-run friction. Users must run docker logs … | grep token before they can fix anything. Kuma and Homarr set up in the browser. Add a first-run claim screen (a one-time setup code), then a session login (roadmap Phase 3).
  6. Notification setup lives in config files. Kuma configures 90+ providers in the UI with a Send test button. Add a "Where should alerts go?" section in Settings with a test button, backed by the planned Apprise bridge.

P1: expected baseline

  1. History ranges (1h, 24h, 7d, 30d) for host metrics and uptime percentages. Beszel and Kuma have them.
  2. More check types in the Add-monitor sheet: TCP port, ping, DNS, TLS certificate expiry, and heartbeat/cron push URLs (Kuma, Gatus, Maintenant). Keep the "paste one thing, fold the rest" form.
  3. Per-container CPU/memory and "image update available" (Beszel, Maintenant, WUD/Diun).
  4. Maintenance windows and per-monitor / per-container silence. Today silence is global only.
  5. Installable PWA with web push. We claim phone-first but rely on Telegram for push; Beszel ships an iOS app.
  6. Read-only shareable status page so family can answer "is Jellyfin down?" themselves (Kuma, Gatus, Maintenant).

P2: differentiators and polish

  1. Auto-heal policy UI: opt in per container, attempts and backoff, a "gave up after 5 tries" state with escalation.
  2. Visual network root-cause chain (this server → gateway → DNS → ISP) once the WAN dependency checks land.
  3. Compact density toggle for Sam; incident CSV export and a shareable outage report.
  4. Multi-host switcher (Phase 4) and i18n (strings are already centralized).

UX fixes seen in the current screens

  • Container names are used as headlines in lowercase ("blog has stopped"). Use "Container blog has stopped" with the name in code style, or the monitor's display name when correlated.
  • "Since your last visit" repeats items already shown in "Needs attention". Group per item and dim what is already listed above.
  • The host sparkline under "Vital signs" has no label or range.
  • On mobile, problem action buttons are indented under the icon column; they should span the card width.

What we take, and what we avoid

Take

  • Uptime Kuma's heartbeat bars: instantly readable history. Sheltie mutes the healthy bars so failures pop.
  • Beszel's restraint: clean tables, few colors, data first.
  • Dozzle's logs-next-to-the-problem: diagnosis belongs in the rescue flow.
  • Homarr's approachability: visual setup, no YAML needed for common tasks.
  • Status-page clarity from Gatus and Kuma: one glance, one answer.
  • Editorial SaaS polish for the website (cream canvas, ink CTAs, saturated feature cards), kept off the dashboard where hue means status.

Avoid

  • Chart walls that require expertise to read (Netdata's strength for experts, a weakness for our primary persona).
  • Long forms up front. Ask for the one thing that matters (the URL) and fold the rest.
  • Color as decoration. When everything is green and blue, red stops meaning something.
  • Silent staleness. A dashboard that shows old data as live is worse than no dashboard.

Sheltie's UX moat

  1. Narrative over metrics. The Overview reads like a short report: what's wrong, what happened, how the vitals look.
  2. Actionable by default. Problems come with the button that fixes them, gated by confirmation and the action token.
  3. Outage honesty. The UI is designed for the moments the network or server is unhealthy, which is exactly when other dashboards go blank.
  4. Phone-first. Most checks happen away from the desk.
  5. A written design contract. DESIGN.md, CODEOWNERS and CI checks keep the experience consistent as contributors join.

Closing the P0 gaps turns the moat from "tells you what's wrong" into "gets you from alert to fixed without leaving the page".

Sources

Edit this page on GitHub