Dashboard Overview

After you log in, the Overview page gives you infrastructure at a glance. If no agents are connected yet, it shows an empty state prompting you to deploy your first agent.

Dashboard picker

A dropdown in the top-right corner of the Overview page lets you choose which dashboards to display, via checkboxes. Only dashboards with something to show ever appear as options, never a "coming soon" placeholder. Your selection is remembered per browser, so everyone on your team can tailor the Overview page to what they actually run without affecting anyone else's view. The available dashboards are described below; Docker and Kubernetes are on by default.

Docker dashboard

• Stat cards: Total Containers, Running, Stopped, and Active Alerts.
• Per-container time-series charts: CPU usage, memory usage, network RX/TX, and disk read/write, sourced from each container's live Docker stats. Network and disk charts are cumulative totals since the container started (the same convention as docker stats), not a per-second rate.
• Network chart: containers per Docker network.

Kubernetes dashboard

Shown when a cluster is connected:

• Stat cards: Nodes (ready / total), Running Pods, Pending Pods, and Namespaces.
• A Pods Ready gauge and a Node Info table (role, status, OS, kernel version, Pod CIDR).
• Cluster-reported charts: API server request rate by verb, Kubernetes services by type, and ingress by namespace. These come from the API server's own metrics, so they only report data while an agent is actually leader-elected against a live Kubernetes API server.
• Node host metrics: per-node CPU and memory usage, network RX/TX, disk IOPS, context switches, and file descriptors, collected directly from each node by the KubeWatch agent, with no extra setup required beyond having the agent installed on your cluster. These still report normally even while the cluster-reported charts above show no data.

Kafka Topics dashboard

Shown once you've configured at least one Kafka cluster for the agent to connect to. The agent talks directly to Kafka's admin API, so there's no separate collector to run:

• Cluster / Topic / Consumer Group filters that narrow every chart on the page.
• Topic partition offsets, a proxy for topic size over time.
• Consumer group total lag, per-partition lag, and committed offsets: the numbers you watch to catch a consumer falling behind before it becomes an incident.

Kubernetes Nodes, Kubernetes Monitoring, and Connected Cluster Nodes dashboards

Three additional Kubernetes-focused views alongside the main Kubernetes dashboard above:

• Kubernetes Nodes: per-node CPU, memory, network, and disk detail, one node at a time via a node selector, so a cluster with many nodes isn't a wall of overlapping lines.
• Kubernetes Monitoring: cluster-wide network, CPU, memory, filesystem, and per-pod usage, aggregated across the whole cluster rather than per-node.
• Connected Cluster Nodes: per-node CPU and memory specifically for clusters you've connected via a kubeconfig rather than an installed agent.

PostgreSQL dashboard

Sessions, transactions, cache hit rate, and buffer usage for a Postgres instance the agent is configured to monitor.

Cloud Costs dashboard

Estimated charges per cloud service, daily and over time, from your cloud provider's real billing data. See Cloud Costs for how this differs from the Cluster Cost & Utilization dashboard below.

Cluster Autoscaler dashboard

Node and pod activity, plus scaling status, for the Kubernetes Cluster Autoscaler running in your cluster, separate from KubeWatch's own Auto Scaling feature.

OpenTelemetry Collector dashboard

Receiver, processor, and exporter throughput for a customer-run OTel Collector, so you can see whether telemetry is actually flowing through it before debugging further downstream.

NGINX Ingress & Gateway API dashboard

Per-ingress request volume, success rate, and latency for classic NGINX Ingress objects, plus, on clusters using the newer Kubernetes Gateway API instead: Gateway and HTTPRoute/GRPCRoute status (which routes are accepted and programmed), and reload health if you're running NGINX Gateway Fabric specifically. Both models are shown side by side. You don't need to pick one to get visibility, and a cluster using only one of the two simply shows an empty section for the other.

Certificates dashboard

X.509 certificate expiry for every Kubernetes Secret KubeWatch can read, with adjustable warning thresholds so you catch a renewal that's about to lapse before it takes something down.

AI Agent Cost & Reliability dashboard

Cost, tool calls, errors, and latency per AI agent session. See the AI Agents page (under AI-Ops in the sidebar) to manage policies and take action on what this dashboard surfaces.

Host Metrics dashboard

Per-host load average, memory/swap, CPU, disk throughput, filesystem usage, and uptime, across Kubernetes nodes and Docker hosts alike, useful when you want one view spanning both kinds of agent.

Cluster Cost & Utilization dashboard

A manual-rate cost estimate for Kubernetes clusters: CPU, RAM, storage, and network egress cost computed from current cluster capacity/usage against $/unit rates you set, plus CPU/RAM utilization gauges and a namespace-by-namespace cost breakdown. This is separate from the Cloud Costs page (real AWS/Azure/GCP billing data) and the Nodes page's per-node cost (real cloud instance list pricing). Use it for bare-metal/local clusters with no cloud billing available, or to model a negotiated rate that differs from list price.

OpenCost / Kubecost {#opencost}

If you already run OpenCost or Kubecost in the cluster, point KubeWatch's agent at it to replace the namespace table's usage-share estimate with its real, requests-aware allocation, including idle (unallocated) cluster capacity. KubeWatch doesn't install either for you, just reads from an instance you already have. The two expose the same allocation data at different API paths, which is what costProvider below selects between:

1. Install OpenCost (see OpenCost's own installation docs) or Kubecost (see Kubecost's own installation docs) in your cluster, if you haven't already.
2. Set openCostEndpoint in your kubewatch-agent Helm values to its in-cluster Service URL. For OpenCost, e.g. http://opencost.opencost.svc.cluster.local:9003; for Kubecost, e.g. http://kubecost-cost-analyzer.kubecost.svc.cluster.local:9090.
3. Set costProvider to opencost (the default) or kubecost to match, and upgrade the release.
4. The Namespace Cost Allocation table switches to real data automatically once the agent's next push includes it. No dashboard configuration is needed.

If openCostEndpoint isn't set, or the configured instance is unreachable, the table keeps using the manual-rate estimate.

Alert count

The Docker dashboard's Active Alerts stat card shows how many alerts are currently firing, with a critical-count subtitle when any are critical severity. For the full list of firing alerts, see the Alerts page in the sidebar.

Navigation

The left sidebar is grouped into collapsible sections, which auto-expand to reveal whichever page you're currently on and otherwise stay collapsed:

• Overview: this page.
• Main: Containers, Clusters, Network, Alerts, Incidents. Logs aren't a separate sidebar page: they live as a tab on each Pod's and Container's own detail page, see Logs.
• Kubernetes: Pods, Nodes, Workloads, Control Plane, Apply Manifest, Secrets & ConfigMaps, Helm Releases, Config, RBAC, CRDs. See Kubernetes for details. Approvals for edits, manifest applies, and Helm releases live at the bottom of their respective pages, not as separate pages.
• Platform: Integrations, Observability (which itself has Metrics, Traffic, Traces, Profiles, OpenTelemetry, Latency, and API Requests as tabs within the page, not separate sidebar items), Load Testing, Auto Scaling, Cloud Costs, Disaster Recovery, Vault.
• AI-Ops: AI Observability, AI Agents, Model Fleet, Advisor, Remediation, Root Cause AI, Edge Fleet. See the AI-Ops Roadmap section for what each of these covers.
• Admin: Users, Audit Trail, Settings, System Health, System Update, Report a Bug.

This list covers the groups relevant to the pages documented here. The sidebar also has Pipelines, Infrastructure, and DataOps groups for other parts of the platform, not covered by this page.

The theme (light/dark) toggle and sign-out are in the sidebar footer, and your organization name appears at the top of the sidebar (your own logo and accent color too, if you've set them under Settings → Branding).

Agent update notifications

If any connected agent is running an older version than the latest release, a dismissible banner appears at the top of every dashboard page linking to the System Health page, where the outdated agent is flagged with an "outdated" badge next to its version. Dismissing it snoozes that notice for 7 days or until a newer version ships, whichever comes first. It reappears after that if the agent is still outdated, and dismissing one version's notice never suppresses a later version's.