CLI Machines Metrics

ed machines metrics

Samples a machine, once or continuously. It is the same collector the app's Machines view drives, fed to the machine on stdin, so nothing is installed there and nothing is left behind.

ed machines metrics <machine> [--json] [--follow] [--interval <seconds>]
                              [--processes <n>]

Arguments

Name	Type / values	Default	What it does
<machine>	string, required	none	Machine name, ssh alias, id or unambiguous prefix.

Options

Name	Type / values	Default	What it does
--json	flag	off	Emit JSON on stdout instead of the human lines.
--follow, -f	flag	off	Keep streaming until interrupted. Also switches --json from one pretty document to one compact document per line.
--interval	integer seconds, greater than 0	2	Seconds between samples when following. Ignored without --follow.
--processes	integer, 0 or more	0	Include this many of the processes each sample carries, out of the thirty at most that the collector sends.
--help, -h	flag	off	Print the help for this command on stdout and exit 0.

The first line is the collector's greeting, carrying the machine's own host name, its OS string and its core count. Each later line is a sample:

$ ed machines metrics studio
studio-mac  macOS 26.6.1  20 cores
cpu   1.0%   mem 5% of 67.0 GB   load 0.12 0.14 0.30   net down 132 B/s up 1.6 KB/s

Without --follow it prints exactly one sample and exits. With --follow the greeting prints once and a sample line is added every --interval seconds until you interrupt it.

--json shape

One object with a host half that never changes and a sample half that does. This is a real document with the core list, the device list and the process list trimmed:

{
  "host": {
    "arch": "x86_64",
    "cores": 20,
    "cpuModel": "Apple M4 Pro",
    "host": "studio-mac",
    "kernel": "7.0.0-28-generic",
    "memTotalKB": 65452140,
    "os": "macOS 26.6.1",
    "osID": "macos",
    "virtual": false
  },
  "sample": {
    "at": "2026-08-08T16:48:25Z",
    "cpu": {
      "corePercent": [0, 0, 1.8, 0, 2.7],
      "stealPercent": 0,
      "totalPercent": 0.9
    },
    "disk": {
      "devices": [
        {
          "busyPercent": 0,
          "name": "disk3",
          "readBps": 0,
          "writeBps": 24576
        }
      ],
      "readBps": 0,
      "writeBps": 24576
    },
    "intervalSeconds": 1,
    "load": [0.12, 0.14, 0.3],
    "memory": {
      "availableKB": 62045152,
      "buffCacheKB": 56620508,
      "swapTotalKB": 8388604,
      "swapUsedKB": 376,
      "totalKB": 65452140,
      "usedKB": 3406988,
      "usedPercent": 5.205311850766072
    },
    "network": {
      "interfaces": [
        {
          "name": "wlo1",
          "rxBps": 316,
          "txBps": 1550,
          "virtual": false
        },
        {
          "name": "docker0",
          "rxBps": 0,
          "txBps": 0,
          "virtual": true
        }
      ],
      "rxBps": 316,
      "txBps": 1550
    },
    "processes": [
      {
        "command": "node /opt/unduck/node_modules/.bin/vite preview",
        "cpuPercent": 0,
        "memPercent": 0.1,
        "name": "MainThread",
        "pid": 1857,
        "rssKB": 91788,
        "user": "pulkit"
      }
    ],
    "tasks": {
      "runnable": 2,
      "total": 1117
    },
    "uptimeSeconds": 33029
  }
}

What the fields mean:

• host.os is what the machine calls itself, from sw_vers, and host.osID is its short id such as macos. host.virtual is the collector's judgement about whether it is a VM.
• sample.at is the sample time, and sample.intervalSeconds is how long the window behind this sample actually was.
• cpu.totalPercent is 0 to 100 across the whole machine, cpu.corePercent has one entry per logical core in core order, and cpu.stealPercent is time the hypervisor took, which is 0 on bare metal.
• Every *KB number is kilobytes and every *Bps number is bytes per second. memory.usedPercent is usedKB over totalKB.
• load is the one, five and fifteen minute load averages, in that order.
• disk.devices is per block device with a busyPercent, and network.interfaces is per interface with a virtual flag that labels bridges and container interfaces. The flag is a label only: the rxBps and txBps totals add up every interface the machine reports except loopback, virtual ones included.
• processes is present even when it is empty, so the key never disappears between runs. With the default --processes 0 it is always []. The collector sends at most thirty processes, the busiest by CPU plus the largest by memory, in no particular order, so --processes trims that list rather than ranking it.

Examples

ed machines metrics studio
ed machines metrics studio --json
ed machines metrics studio --processes 20
ed machines metrics studio --follow --interval 5 --json | jq -c '{cpu: .sample.cpu.totalPercent}'

Behaviour notes

Nothing is written locally and nothing is installed remotely. ed opens the shared connection, runs sh -s -- --once or sh -s -- --stream -i <interval> there, and pipes the collector script into that shell's stdin. The script needs a POSIX shell and awk and nothing else.

The collector's own stderr is discarded, so a warning on the machine never pollutes the report.

Failures, with their codes:

• an unknown or ambiguous machine name exits 3
• a machine that cannot be reached exits 4
• a machine that connects but never emits a sample exits 4 with <name> did not report metrics and the hint that the collector needs a POSIX shell and awk
• --interval 0 or a negative interval exits 2 with --interval must be greater than zero, and --processes=-1 exits 2 with --processes cannot be negative; both are checked before the machine is dialled
• a build with the collector script missing exits 1

Write a negative process count as --processes=-1. Spelled --processes -1 the parser reads it as a missing value and exits 2 for that reason instead.

--json --follow writes one compact document per line, forever, repeating the whole host object on every line so each line stands alone for jq -c, head or a pipe. Without --follow you get a single pretty document.

The collector also emits a slower filesystem record. ed machines metrics decodes and discards it because filesystem data has its own Machines view.

ed system stats is the same report for the Mac you are typing on, in the same shape, so a script can treat local and remote the same way.

Where to go next

• ed machines, the rest of this group
• All ed commands