Hypervisors: vSphere, Proxmox VE, Hyper-V, KVM and VirtualBox
The state of hypervisors and their VMs, read-only, next to the agents installed inside the VMs.
VMware vSphere (ESXi and vCenter)
No agent goes on an ESXi host: an agent of your fleet queries the vSphere API with an account holding the ReadOnly role.
integrations:
vsphere:
interval: 60s
targets:
- name: vcenter
url: https://vcenter.example/sdk
username: hypha@vsphere.local
password: "…"
thumbprint: "AA:BB:…" # SHA-1 fingerprint of a self-signed certificate
# insecure: true # avoid: accepts any certificate
tags: {site: paris}
What HYPHA collects:
- ESXi hosts: state, alarms, maintenance mode, CPU, memory, running VMs, uptime (
vsphere.host.*); - datastores: accessibility and usage (
vsphere.datastore.*); - VMs: power, state, guest heartbeat, CPU, memory, age of the oldest snapshot (
vsphere.vm.*); - vSphere events.
ESXi hosts and VMs become resources in Infrastructure. A VM that also runs a HYPHA agent is matched to its host by serial number ("VMware-…"): one line, and its page shows what vSphere says about it.
Proxmox VE
Two ways to read the cluster:
- On a Proxmox node: the agent uses
pveshas root, without an account, in a separate, bounded systemd unit. Only the first online node reads the cluster: the other agents stay quiet, without duplicates. - Remotely: the API with a token holding the PVEAuditor role.
integrations:
proxmox:
url: https://pve.example:8006
token_id: "hypha@pve!readonly"
token_secret: "…"
ca_file: /etc/hypha/pve-ca.pem
Checks: cluster quorum, node offline, storage unavailable, HA guest out of its requested state, last backup (vzdump) of each node failing, cleared as soon as the next one succeeds. Metrics for nodes, storages and guests (VMs and containers). Failed tasks become events.
Hyper-V
On Windows, the agent reads Get-VM when the Hyper-V service is present; no setting needed.
hyperv.vmcheck per VM: critical state (storage unreachable), paused, stopped while its automatic start is "always", guest heartbeat lost.hyperv.vm.replicationcheck: replication health.- Metrics: CPU, assigned and demanded memory, vCPUs, uptime, number of checkpoints and age of the oldest (a forgotten AVHDX file grows endlessly).
KVM and libvirt
On the host, virsh read-only (qemu:///system by default, set by uri).
- Domains: state and reason, CPU, vCPUs, memory, disks, network (
libvirt.domain.*). - Checks: crashed domain, paused on an I/O error (full pool, the classic case), stopped while autostart is on; storage pool inactive or inaccessible.
- Pools: usage and available space (
libvirt.pool.*).
VirtualBox
The host's agent collects the declared and running VMs, and reports an autostart VM that is stopped (virtualbox.vm.running). The user setting follows a single account.
Monitor templates
The Virtualization family of the catalogue: vSphere connector failing, ESXi host faulty, datastore inaccessible or almost full, vSphere VM faulty, old snapshot; Proxmox unreachable, quorum lost, node offline, storage unavailable or almost full, HA guest out of its state, backup failing; Hyper-V VM faulty, old checkpoint, degraded replication; KVM VM faulty, pool inactive or almost full.
Updated: 19 November 2026