The swap chip's tooltip reads swap 16.7GB of 18.0GB · pressure normal — close sessions (save a list first) or quit heavy apps; macOS only clears swap on a restart, and in context — a chip that sits beside ● 23 live, in an app about Claude Code sessions, with advice about closing sessions — it reads as Claude Code is using 16.7 of 18 GB. That is not what either number means, and the misreading happened to the person who asked for the feature (2026-09-08).
What the numbers actually are
Both come from sysctl -n vm.swapusage, which is machine-wide and has no per-process concept at all:
used (16.7GB, the figure on the chip) is how much of the swap file currently holds paged-out data from every process on the Mac.total (18.0GB) is the swap file's current size. macOS grows it on demand, so it is not a limit, and it is not physical RAM (32GB on the reference machine).
What Claude Code's actual share was
Per-process swap is measurable — vmmap --summary <pid> reports a SWAPPED column on its TOTAL row — so the real attribution on the same machine, minutes later:
|
|
27 claude processes, swap |
4.24 GB |
| machine swap in use |
16.46 GB |
| Claude Code's share |
26% |
the same processes, resident (ps RSS) |
5.43 GB |
| largest single process |
325 MB swapped |
So roughly a quarter, against an intuition of nearly all of it. (Summing per-process figures double-counts shared pages, so treat 4.24 GB as an upper bound.)
Proposal
Reword the tooltip so the scope is stated rather than inferred. Something like: swap 16.7GB of the machine's 18.0GB swap file (every app, not just Claude) · pressure normal — …. The chip's own label can stay swap 16.7GB; it is the tooltip that has room to say whose swap it is.
Not proposed: showing Claude Code's share on the chip. One vmmap --summary call was measured at 1.8 seconds; 27 live processes would be about a minute per refresh, which is far past what this chip should cost. The machine-wide figure is the cheap one (sysctl on the same refresh as the process table) and is still the number that says "this machine is in trouble" first — which is why it was added.
Follows #152, which shipped the chip.
🤖 On behalf of @grimmerk — generated with Claude Code
The swap chip's tooltip reads
swap 16.7GB of 18.0GB · pressure normal — close sessions (save a list first) or quit heavy apps; macOS only clears swap on a restart, and in context — a chip that sits beside● 23 live, in an app about Claude Code sessions, with advice about closing sessions — it reads as Claude Code is using 16.7 of 18 GB. That is not what either number means, and the misreading happened to the person who asked for the feature (2026-09-08).What the numbers actually are
Both come from
sysctl -n vm.swapusage, which is machine-wide and has no per-process concept at all:used(16.7GB, the figure on the chip) is how much of the swap file currently holds paged-out data from every process on the Mac.total(18.0GB) is the swap file's current size. macOS grows it on demand, so it is not a limit, and it is not physical RAM (32GB on the reference machine).What Claude Code's actual share was
Per-process swap is measurable —
vmmap --summary <pid>reports aSWAPPEDcolumn on itsTOTALrow — so the real attribution on the same machine, minutes later:claudeprocesses, swappsRSS)So roughly a quarter, against an intuition of nearly all of it. (Summing per-process figures double-counts shared pages, so treat 4.24 GB as an upper bound.)
Proposal
Reword the tooltip so the scope is stated rather than inferred. Something like:
swap 16.7GB of the machine's 18.0GB swap file (every app, not just Claude) · pressure normal — …. The chip's own label can stayswap 16.7GB; it is the tooltip that has room to say whose swap it is.Not proposed: showing Claude Code's share on the chip. One
vmmap --summarycall was measured at 1.8 seconds; 27 live processes would be about a minute per refresh, which is far past what this chip should cost. The machine-wide figure is the cheap one (sysctlon the same refresh as the process table) and is still the number that says "this machine is in trouble" first — which is why it was added.Follows #152, which shipped the chip.
🤖 On behalf of @grimmerk — generated with Claude Code