Host Metrics
Live CPU, memory, disk, network and GPU stats, plus tools to manage your hosts.
- CPU, memory, disk, network, temperature, NVIDIA and AMD GPUs, processes, ports and logins
- History charts for each host
- Run health checks on each host
- Health checks for your hosts and services
- A layout you can rearrange per host
- Live stats in the terminal toolbar
- A live metrics panel beside any SSH terminal
- A dashboard card with the hosts working hardest
Host Metrics shows live stats for your Linux hosts: CPU, memory, disk, network, temperature, NVIDIA and AMD GPUs, processes, open ports and logins. It keeps history for charts and runs health checks.
Everything is read over SSH. There is no agent to install.
Open it
Pick Host Metrics from a host's menu. Metrics are on for every host at first. Turn them off for a host in its Host Metrics settings.
Host Metrics is built for Linux. It reads /proc and standard tools, so Debian, Ubuntu, Fedora, Arch, Alpine and most others work. macOS and Windows hosts get CPU, memory, disk, network, uptime and system info, but not the other cards or the managers. BSD isn't supported.
Layout
The page is made of cards on a 12 column grid. Drag them to move them, resize them, and turn cards on or off. The layout is saved per host.
Choose Customise to move cards and resize them by their left, right or bottom edge or the bottom right corner. A dashed outline shows where a card will land, and only the cards in its way move down. Widths are whole columns and heights snap to nearby cards. You can also move and resize with the arrow keys, change each card's presentation settings, and undo or redo changes with Ctrl+Z and Ctrl+Y.
In the dashboard's edit mode, Dashboard Templates saves your current layout as one of up to 50 private templates and copies it once to up to 100 hosts you can access. It replaces only your own layouts on those hosts; later template changes or deletion do not affect them. On hosts you own, applying a template also updates the enabled widgets used by the mobile view; shared-host settings and sampling intervals stay unchanged.
Export and import
In edit mode, Export / Import saves your layouts, health checks and dashboard templates as a JSON file. Tick Include history to add metric samples and health results of the last 30 days. Import reads the same file. Hosts are matched by address, port and user, so the file also works on another server. Hosts that are not found or that you cannot connect to are skipped. Existing layouts, checks and templates are kept unless you choose to replace them. History that is already stored is never added twice. Metric samples are only added for hosts you can edit, since everyone who can see the host shares them.
The same is available as an API: GET /plugin-api/host-metrics/export?history=1&from=&to=&hostIds= and POST /plugin-api/host-metrics/import with { "data": <file>, "overwrite": false, "history": true }. Both are logged in the audit log.
The cards: CPU, memory, swap, disk, network, temperature, GPU, uptime, system info, top processes, listening ports, recent logins and firewall status.
Swap Usage
On Linux hosts, add it with Customise > Add card > Swap Usage. It shows used, free and total swap, read from /proc/meminfo without sudo. It is also a host page tile and a homepage history chart metric. It is off by default.
"No swap configured" means the host has no swap. "Swap metrics unavailable" means the host is not Linux or the numbers could not be read. History from before this version shows as gaps.
The terminal toolbar shows small CPU, memory and disk bars for the host you are in.
Metrics panel beside a terminal
Open the Host Metrics button on the right edge of an SSH terminal, or Metrics panel in the tab's menu, to watch the host while you type. It shows CPU, memory, disk, temperature, uptime and the busiest processes, and updates on the host's metrics interval. Drag its edge to resize it and use the arrow in its header to move it.
In Settings, Host Metrics you can open the panel with every new terminal, pick its side and size, and hide the edge button. Turn off Show metrics panel in terminals in the host's Host Metrics section to keep it from opening there.
Busiest hosts card
Add the Busiest hosts card in the dashboard's edit mode. It ranks hosts by CPU or memory, whichever is higher, using samples Termix already collected, so it never opens new SSH connections. A host shows once something has collected its metrics, like its metrics tab or the host status card.
History
The clock button on a card opens its history chart: the last 1 hour, 6 hours, 24 hours, 7 days, 30 days, or a range you pick.
Samples are taken while someone is looking at a host, every 30 seconds by default. Admins set Metrics interval and History retention, 7 days by default, in Settings, Host Metrics. A host can set its own interval.
Live updates
While a Host Metrics tab is open and visible, CPU, memory, network, the process list and NVIDIA or AMD GPUs refresh every 2 seconds, and the CPU card shows a bar for each core. Pick Off, 1, 2 or 5 seconds under Live update rate in your Host Metrics settings. Admins control this with Live updates (on or off) and Live update minimum (default 2 seconds). Each host is sampled at most once per minimum, however many users or tabs are watching, and a faster user choice is slowed to it. This is separate from the Metrics interval, which only applies to regular sampling and history. Live updates also stop for hosts with Collect metrics turned off. Live updates read /proc on the host over SSH, so they only work on Linux. Each update runs one short awk command on the host, and updates pause while the tab is hidden. On hosts with a GPU, the same nvidia-smi and amdgpu readings as regular sampling also run with the updates, at most every 2 seconds, and are stopped after 5 seconds when the host has timeout. Hosts without a GPU skip them. The CPU and memory bars in the terminal toolbar follow the same updates, and dashboard rows and homepage widgets use them too, but never faster than every 5 seconds and only while a host is online and shows a live field. Process search, sort and tree cover the busiest 60 processes. Other cards, such as disk, ports and the firewall, keep the regular sampling.
CPU, memory, physical disks and adapters
These cards describe the remote host, not the computer running your browser. They add Task Manager-like data without copying its layout:
- CPU: utilization and logical-processor histories, model, reported current/base/maximum frequency, sockets and core counts, virtualization information, caches, process/thread/handle counts and uptime.
- Memory: used and available composition, compressed memory, committed memory and its limit, cache and kernel pools where supported, plus installed capacity and available module information. Cached memory overlaps other categories, so it is listed separately rather than stacked into the composition bar.
- Disk: physical-device active time, read/write rates and histories, average transfer latency, capacity and available device metadata. Filesystem space usage remains separate: a filesystem that is 90% full does not mean its physical disk is 90% active.
- Network: separate send/receive histories for each interface, adapter identity, IPv4/IPv6 addresses, interface-specific DNS and reported link information.
Use each card's existing presentation settings under Customise to show or hide the new groups and adjust chart size. These are personal display choices: hiding details does not change collection for other viewers. Existing host/admin metric and widget controls gate the additional CPU, memory, disk and network queries.
The new histories hold at most 60 points in the open card and reset on a host change. Linux CPU and network histories can follow live updates; hardware, physical-disk counters and the other platforms follow regular polling. These detailed series are not added to the historical database: the time-range tabs still show the existing aggregate history. First samples, counter resets and unavailable sources produce gaps, not made-up zero activity.
Platform sources and limits
Unsupported or unreadable fields show Unavailable. No elevation, firmware dump, extra agent or package installation is attempted. Additional collection is read-only, with a 15-second command timeout, and failure leaves baseline metrics available.
| Platform | Sources and limits |
|---|---|
| Linux | CPU topology/cache comes from lscpu, frequency and utilization from /proc and sysfs. Virtualization describes hardware support, not a Windows firmware-enabled flag. Handles means system-wide allocated open file handles, not Windows process handles. Memory uses /proc/meminfo: committed virtual memory, page cache plus buffers, and reclaimable/unreclaimable slab are explicitly labelled Linux equivalents. Compressed RAM, Windows pools, DIMM inventory and hardware-reserved RAM are unavailable. Disk rates, active time and latency use physical-device lsblk identity and /proc/diskstats deltas, excluding partitions and logical volumes. Formatted capacity, system/pagefile attribution are unavailable rather than inferred from mount paths. Network model is the sysfs driver identity when available; DNS requires interface-specific resolvectl output. |
| Windows | Read-only CIM and Storage/NetAdapter commands provide processor, RAM, physical-disk and adapter data. Current clock and reported maximum clock are separate; an actual base frequency and compressed RAM counter are unavailable from these sources. Cached memory sums standby lists and the modified list, not the system-cache working set. Hardware-reserved RAM is estimated as installed module capacity minus OS-visible RAM. Module speed is the configured MHz value reported by CIM. Disk active time uses idle counters; submillisecond latency uses raw counter deltas and requires two samples with completed operations. SSD/HDD media requires matching physical-device identity. Formatted capacity and system/pagefile flags are provided only when their disk/partition sources can be read. |
| macOS | sysctl, vm_stat, diskutil and ifconfig provide available topology, RAM, physical-device and interface details. Compressor and wired bytes use the actual VM page size; cache means file-backed pages. Per-logical-processor utilization, physical-disk activity/rates/latency, base frequency, process handles, committed memory/pools, DIMM inventory and interface-specific DNS remain unavailable with these commands. CPU frequency fields can also be unavailable, especially on Apple Silicon. Disk/partition roles and formatted capacity are not guessed. |
MikroTik RouterOS v7 has its own shell, so Termix reads it with RouterOS commands instead: CPU load, memory, flash storage, uptime, board temperature, and traffic per interface. Processes, logins, ports, firewall, GPU and the managers are not available on RouterOS.
Permission restrictions, missing commands and unsupported counters can reduce this coverage on any platform. A zero rate is shown only for a valid sampled counter, not as a substitute for an unavailable value.
Windows hardware inventory is cached per host for up to five minutes; utilization, clocks, memory use, disk counters and adapter state are still sampled each time. Charts ease their newest point over about 800 ms using a shared animation clock, and update immediately when reduced motion is enabled. This does not change the host's sampling interval.
Managers
The Health Checks card is built in.
The Firewall card shows whether the firewall is on and its rules. It only reads. To change rules, install the Firewall plugin: it fills the Firewall manager card here and adds a Firewall tab for each host.
Other plugins fill the rest of the manager cards when they are installed: Process Manager fills Process Inspector, Services fills Services, Timers and Log Viewer, WireGuard fills WireGuard, Packages fills Packages, Cron fills Cron Jobs, Users fills Users and Certificates fills SSL Certificates.
The card detects the firewall that owns the rules. Active UFW takes precedence over its underlying iptables or nftables rules. Reads use the saved Sudo Password when needed, without trying sudo on every poll when no password is saved. An unreadable or unrecognised state shows as unknown.
On a shared host, seeing the managers only needs connect access. Changing a host needs edit access, and the buttons are hidden without it.
Health checks
Health checks test a URL or a TCP port from the host, and show when one fails. Automations can act on a check changing.
Checks run in the background on their own schedule, every 5 minutes by default. Set Check every (minutes) when you edit the checks, down to 1 minute. They also run when you open the card or press Run. Scheduled checks run as the user who saved them, and are skipped while metrics are off for the host, while that user can no longer reach it, or when the host's login needs a code typed in.
When a check starts failing or comes back, the user who saved it gets an alert in Alerts.
Script checks
A Script check runs your own command on the host and shows what it returns. Pick how to read it:
- Pass on exit code 0: passes when the command succeeds, for example
systemctl is-active auditd. - Number: reads the last number the command prints. Set Warn at and Fail at to color it, and Low is bad when a low
value is the problem, like free space. For example
df --output=pcent /var/log/audit | tail -1with unit%. - Show output: shows the first line the command prints, and fails on a non zero exit code.
Turn on Show on dashboard for a check to see its latest result in the host's row of the dashboard's host list, next to CPU, RAM and disk. If any of those checks warns or fails, the host's status badge shows Warning or Failing instead of Online. The dashboard only reads results that were already stored, so checks run on their own schedule or when you press Run. Each user picks their own dashboard checks.
The command runs as the SSH user, with a 15 second limit. A number check fails only past Fail at, so an alert is sent when it crosses that line.
What the SSH user needs
Most stats come from files anyone can read. A few need more:
| Stat | Needs |
|---|---|
| Failed logins | Read access to /var/log/auth.log or /var/log/secure. |
| Process names on ports | sudo, for ss -tulpn. |
| Firewall rules | sudo, for ufw, iptables-save or nft, when required. |
| GPU | nvidia-smi for NVIDIA, the amdgpu driver for AMD. |
Without them, those parts are empty. The rest still works.
With other plugins
- Automations can run when a metric crosses a line or a health check changes.
- Homepage gets a Metrics Chart widget.
- The dashboard's host status card shows each host's numbers.
Who can use it is set by the host-metrics.use permission. Admins and users have it at first.
Environment variables
This plugin reads some environment variables on the Termix server. See the reference for the list.