Your Mac, across
the menu bar.

CPU

One bar per core, live. On Apple silicon the performance and efficiency cores are split into their own groups and can carry their own colours, so you can see at a glance whether real work is happening or the machine is just ticking over.

The popover adds the load history and the processes actually burning the time.

• Menu barPer-core bars, a single bar, or a line graph
• PopoverHistory, per-core groups, top eight processes
• ReadsHost CPU load ticks — no subprocess for the meter

Memory

How much memory is actually in use, as a single bar you can read without stopping what you are doing. No "free RAM" theatre — the figure tracks real pressure.

Open it to see which applications are holding the memory, largest first.

• Menu barSingle bar, or a line graph
• PopoverUsed, wired, compressed, and the top eight apps

Network

Throughput up and down, as a scrolling graph with the rates beside it. Automatic interface mode follows whatever is actually carrying traffic, so a VPN that hops between tunnel interfaces is tracked without you editing a list.

The popover breaks the traffic down by application, so you can see which one is doing the talking.

• Menu barGraph, rates, or both
• InterfacesAutomatic, or a set you choose
• PopoverPer-app up and down rates

GPU

Device utilisation read from the accelerator itself, using time residency rather than an instantaneous sample — so brief bursts are not missed.

New in 1.1: the popover lists the apps actually working the GPU, ranked by real GPU time. The figure is milliseconds of GPU time per second, not a percentage, because a Mac running several command queues at once can genuinely bill more than one GPU-second per second. A percentage would read "140%" and mean nothing.

• Menu barSingle bar
• PopoverHistory, renderer and tiler split, memory in use, top apps
• ReadsIOKit accelerator statistics — no helper, no permission

Spectrum

A real analyser on the system output mix — 1 to 48 logarithmic bands from 32 Hz to 20 kHz, verified against test tones. It is the module people install MenYew for.

Five ways to draw it, and colour themes modelled on gear that actually sat in hi-fi racks. Try the styles:

New in 1.1: the popover lists the apps sending audio out. A playing app is solid and sorts to the top; one that stops greys out for fifteen seconds, then leaves — so a track change does not make rows flicker.

• StylesBars, LED blocks, analyser trace, VU needle, waterfall
• Themes1980s, Pioneer, Technics, Amber, or your own colour
• ChannelsMono or stereo, with optional peak hold
• PermissionNone for output. The optional mic spectrum asks once

Levels

Two loudness meters in one module. Loudness is broadcast-standard LUFS, K-weighted to ITU-R BS.1770, measured on the system output — for the question "am I at −16 LUFS?" that streaming and podcast work keeps asking.

Room Level points the same engine at the microphone for an approximate room sound level in dB, with a calibration offset you can nudge. Quiet enough to record?

• ModesLoudness (LUFS) · Room Level (SPL)
• LoudnessMomentary, short-term, sample peak. macOS 14.2+
• Room LevelUncalibrated dB with an adjustable offset. Needs the mic

Musician

Tuner listens to the microphone and shows the nearest note with how many cents you are sharp or flat, plus an in-tune band. A stability control trades responsiveness against a steadier reading, so the note stops dancing.

Key detector listens to the output instead and names the musical key of whatever is playing, using a chromagram and Krumhansl-Schmuckler profiles.

• ModesTuner (mic) · Key detector (output)
• TunerNote, octave, cents, and a menu-bar cents gauge
• KeyKey and confidence. macOS 14.2+

Hum Hunter

Something in the room is humming. This finds its single dominant frequency, names the note, and guesses the culprit — a 60 Hz ground loop, coil whine, a fan harmonic.

It also generates a matching tone so you can pinpoint the source by ear, and pink-noise masking when you would rather just cover it. It says plainly that cancelling a room through speakers is not physically possible, rather than pretending otherwise.

• FindsDominant frequency, nearest note, likely source
• ToolsMatching tone, antiphase tone for headphones, pink-noise mask
• NeedsMicrophone access

Claude & Codex

Token consumption per account, read straight from the config folders on disk. Point it at as many accounts as you use and each gets its own bar.

Counted honestly: cache-read tokens are excluded, because the cached prompt is re-read every single turn and counting it turns a real 75 million into billions.

• TracksClaude Code and Codex, multiple accounts each
• CountsInput, output and cache creation — once each
• ReadsLocal session files. No API keys, no network