The phone never talks to Prometheus, Alertmanager or Hubble. Your Kubyl Server does, using the cluster's own access, and the app shows the result. Each view appears only if the server has it (metrics.charts, prometheus, alerts.rules, netflow…).
Charts
Line charts have a range picker (15m 1h 6h 24h 7d), touch scrub with a tooltip, a legend you can toggle, gaps where data is missing, and the desktop's units. They appear on the Overview (CPU and memory with top-N breakdowns plus “other”), on pods, workloads, nodes, namespaces and PVCs. If the server finds no source it shows Connect Prometheus for charts with Look again (re-detect).
Prometheus
- Explorer (More, Prometheus): five tabs. Overview; Query (Table or Graph, Execute, history that stays on the phone and is never sent, examples, completion of metrics, functions, labels and label values from the server); Target health (pools, filter, chips, a target page); Rule health (chips, rule page); Service discovery (pools, a pool page, Show more).
- A server selector when several Prometheus servers are found, Open web UI over the bridge, a sign-in panel for servers behind basic auth, and calm blocking states (detecting, not found, forbidden).
- Older servers without the explorer show the plain Prometheus query page: PromQL with completion, examples and recent queries, as a table or a line chart.
- Query results and flow data are kept in memory only.
Alerts
- Tabs: Alerts, Silences and Rules, with counts and a summary line (counts, pending, silenced, heartbeat, failing rules, when it was checked). Header chips show each Alertmanager (version, peers, via which route) and Open Alertmanager UI; Refresh and Look again are there too.
- Alert page: summary, labels, annotations, runbook, the pod or node it points at, the defining rule and Acknowledge (creates a short silence the server defines). Silenced alerts sort last, pending ones pulse.
- Silences: filter, New silence…, Matches, Starts, Ends, +1h and +4h, Edit…, Recreate…, Expire…; expired ones from the last 24 hours sit behind a header. The editor takes a duration (default 2 hours) or an end time, shows a live match preview, and keeps matchers locked to the cluster. A comment is required.
- Rules: summary, groups with a link to the PrometheusRule, cards with state, health,
for, evaluation and last error, a rule page with Show its alerts, Copy and Edit PrometheusRule. - Not yet: a matcher-syntax filter on the Alerts tab, resolved history, a 24-hour timeline, signing in to Alertmanagers behind auth, and undo toasts.
Network flows
More, Network flows reads Cilium Hubble, Whisker or NetObserv through the server, when it finds one. The header shows the backend (version, nodes), a live chip, the window (15m or 1h; the server keeps one hour) and Pause. Flows has the desktop's filter language (parsed by the server, errors in red, completion, term and verdict chips with counts) and cards with verdict, direction, ends, protocol or L7 summary, bytes and policy. A flow page shows policy, source and destination (tap to filter or open the pod), L7 details, counters and Copy as hubble observe. Topology is a force-directed graph by namespace or workload with edges per verdict and sheets with Show flows. The query is polled every 5 seconds. It has been tested against a fake backend only, since no flow backend ran in the test cluster. A Network flows section also appears on pods, workloads, Services and namespaces.