Skip to content
OpsHero
Book a consultation
Beta

Monitoring that arrives with the cluster

Every cluster on the OpsHero Platform comes with metrics, logs, dashboards and alerts already wired – for Kubernetes, for GPUs, for model serving and for the agents and coding tools your team uses.

Not for you if
  • You would rather wire metrics, logs, dashboards and alerts by hand for every new cluster.
  • Your clusters, GPUs, model serving and agents can each sit in a different tool.
  • Everyone can share one admin login to the dashboards.
The Observe view in the OpsHero Portal: nine dashboards grouped as AI (agents, a coding assistant, inference, NVIDIA GPU and DCGM exporter, sandboxes), Cluster (logs, resources) and General, each tagged with its data source and opened in Grafana with the viewer's own permissions.
The Observe view. Each dashboard opens in Grafana as you, with your own permissions – not through a shared admin login.

An incident does not drive the first dashboard. The dashboards and alerts are configured on Day 0

  • Dashboards and alert rules arrive with the cluster, before the first workload, so the first incident lands on a screen that already exists.
  • AI usage is visible from the first call: which agent, which model, which tool and how many tokens, long before the invoice.
  • GPU, application and model metrics sit side by side from the start, so a slow answer and the hot GPU behind it show up together.

Dashboards from the cluster to the agent, already wired

Cluster & Workload

Cluster health and logs, metrics and traces for every workload. Dashboards: Logs, Resources. Grafana for the browser, MCP for agents.

GPU

Utilisation, memory, power and temperature per card. Dashboards: NVIDIA GPU, DCGM exporter.

Inference

Requests, tokens, latency and which model served them. Dashboard: Inference.

AI usage

Which agents and tools ran, which models they called, which tools they used and what it cost in tokens. Dashboards: Agents, coding assistant.

Open a dashboard and the answer is already on it

GPU

AI usage

Cost detail

1 of 3: Utilisation, memory, clock and PCIe for every GPU on a node

GPU

Utilisation, memory, clock and PCIe for every GPU on a node

The Observe view opening the Machine Hardware dashboard in Grafana for one node: utilisation, VRAM used, SM clock and PCIe throughput for each of its eight GPUs over the last twelve hours.
The Machine Hardware dashboard for one node with eight GPUs, over twelve hours. The folder name is blurred.

There from the day the cluster is

Installed with the cluster. No separate project. A new cluster gets the dashboards and alert rules it needs.

One protocol. Everything reports over OpenTelemetry, including the coding assistants and agents your team runs, so AI usage lands next to infrastructure metrics.

Your permissions. Dashboards open with the viewer's identity.

Your SIEM. Connect to your SIEM or forward the telemetry to your system of choice. Possible with a simple config change.

See your own cluster's dashboards, populated, on a call

Book a walkthrough

Where to go next

Sovereign CloudYour sovereign cloud, run end to end by our engineers.Sovereign AISovereign Inference as a Service.Monitoring and Logs ManagementLogs you can search, alerts that mean something, and a bill that stops growing faster than your estate.
Observability: Clusters, GPUs and AI Agents in One View – OpsHero