You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fast cache recycling no longer counts as pressure unless fast memory is actually short
(for example during large file copies with lots of RAM free).
When the state is high because of virtual memory (commit near its limit), the dashboard
says so: "Virtual memory is nearly full · RAM cleaning won't help".
RAM-only actions no longer run for it.
Virtual memory everywhere:
A dashboard row shows commit against the commit limit (RAM + page file) and page-file use.
Each action in Activity shows how much of the "freed" RAM went to the page file.
The Optimize and Deep clean dialogs say so up front.
One-click Quit offers. When memory stays tight, MemManager offers to quit the idle app
holding the most memory, with a notification and a Quit button. It never quits on its own;
each app is offered at most once every 6 hours, and the offer can be turned off in Settings.
Clear about what it can do:
The Activity page states what MemManager can do on this Mac. macOS manages memory itself;
Nudge needs the helper, which needs a Developer-ID-signed build.
The "Now" line reads "macOS is handling memory".
Upgrading
From 0.2.0 or 0.3.0: MemManager offers this update by itself within a day of its release;
to get it sooner, use Settings → Updates → Check now. It installs in place and keeps your
settings.
From 0.1.0: install this version once by hand.
Known limitations
Unsigned builds:
Windows SmartScreen may warn.
On macOS, right-click → Open the first time. Nudge stays unavailable until builds are
Developer-ID signed.
Not yet hand-tested: the dashboards and the Activity view are verified headless in CI only.
Install
Windows 10/11 (x64 or ARM64): run MemManager-0.4.0-windows-setup.exe. Keep "Start at login with memory-cleaning privileges" checked so MemManager can manage memory without asking for admin each time. Uninstall from Settings → Apps.
macOS 13+: open MemManager-0.4.0-macos-universal.dmg and drag MemManager to Applications. On first launch, right-click → Open (the build is not notarized yet).
Portable Windows zips and a macOS zip are attached as well.
Measured memory reduction
Each row creates real bloat on a GitHub-hosted runner, runs the action through the same code path the app uses, and measures memory freed plus the cost afterwards. Numbers vary by machine and workload.
Windows (CI runner, 15.99 GiB RAM, build 26100)
Bloat
Action
Freed
% of RAM
Cost afterwards
Idle apps holding private memory
Selective trim of idle apps (tier 4)
2.00 GiB
12.5%
re-touch 528 ms vs 12 ms warm; 0 hard page faults
Modified (dirty) pages from trimmed apps
Flush modified list (tier 4)
1.84 GiB
11.5%
pagefile writes; pages become reusable cache
File cache after reading a large file
Clear low-priority cache (tier 2)
0.00 GiB
0.0%
none: only cache Windows was about to discard
File cache after reading a large file
Clear standby cache (tier 3)
6.41 GiB
40.1%
re-reading the file took 16810 ms vs 455 ms from cache
All of the above at once
Deep clean (manual)
5.29 GiB
33.1%
624778 hard faults in the next 10 s (baseline 782); apps re-touch 13496 ms
Identical pages across the system
Combine pages (tier 5)
0.02 GiB
0.1%
CPU scan only
macOS (CI runner, 7.00 GiB RAM)
Bloat
Action
Freed
% of RAM
Cost afterwards
Volatile purgeable memory (app caches)
Nudge (helper)
0.70 GiB
10.0%
apps regenerate purged caches on next use
App holding a cache, cooperative
Nudge (helper)
0.70 GiB
10.0%
the app rebuilds its cache when needed
File cache after reading a large file
Purge disk cache (manual, helper)
3.06 GiB
43.7%
re-reading the file took 1539 ms vs 88 ms from cache
A new Activity view lists every action and warning in plain words: what, why (the
triggering reading), which apps, and the measured result.
Actions appear as dots on the dashboard graph.
"Now:" and "Last:" status lines.
"trimmed"/"quieted" badges on apps (Windows).
Leaner dashboard: detail lines moved into Activity; the cache caption shows only with the
one-time explainer.
Lower idle cost: app rows, graph history and the activity list are built only while the
flyout or popover is open.
App icon for the exe, installer and macOS bundle, generated by scripts/make_icons.py.
Upgrading
From 0.2.0: MemManager offers this update by itself within a day of its release; to get it
sooner, use Settings → Updates → Check now. It installs in place and keeps your settings.
From 0.1.0: install this version once by hand.
Known limitations
Unsigned builds:
Windows SmartScreen may warn.
On macOS, right-click → Open the first time.
Not yet hand-tested: the dashboards and the Activity view are verified headless in CI only.
Windows only: the cache-hit line and the app badges.
Install
Windows 10/11 (x64 or ARM64): run MemManager-0.3.0-windows-setup.exe. Keep "Start at login with memory-cleaning privileges" checked so MemManager can manage memory without asking for admin each time. Uninstall from Settings → Apps.
macOS 13+: open MemManager-0.3.0-macos-universal.dmg and drag MemManager to Applications. On first launch, right-click → Open (the build is not notarized yet).
Portable Windows zips and a macOS zip are attached as well.
Measured memory reduction
Each row creates real bloat on a GitHub-hosted runner, runs the action through the same code path the app uses, and measures memory freed plus the cost afterwards. Numbers vary by machine and workload.
Windows (CI runner, 15.99 GiB RAM, build 26100)
Bloat
Action
Freed
% of RAM
Cost afterwards
Idle apps holding private memory
Selective trim of idle apps (tier 4)
2.00 GiB
12.5%
re-touch 516 ms vs 7 ms warm; 1 hard page faults
Modified (dirty) pages from trimmed apps
Flush modified list (tier 4)
1.58 GiB
9.9%
pagefile writes; pages become reusable cache
File cache after reading a large file
Clear low-priority cache (tier 2)
0.00 GiB
0.0%
none: only cache Windows was about to discard
File cache after reading a large file
Clear standby cache (tier 3)
6.01 GiB
37.5%
re-reading the file took 4929 ms vs 263 ms from cache
All of the above at once
Deep clean (manual)
5.27 GiB
32.9%
606746 hard faults in the next 10 s (baseline 1245); apps re-touch 13099 ms
Identical pages across the system
Combine pages (tier 5)
0.01 GiB
0.1%
CPU scan only
macOS (CI runner, 7.00 GiB RAM)
Bloat
Action
Freed
% of RAM
Cost afterwards
Volatile purgeable memory (app caches)
Nudge (helper)
0.70 GiB
10.0%
apps regenerate purged caches on next use
App holding a cache, cooperative
Nudge (helper)
0.70 GiB
10.0%
the app rebuilds its cache when needed
File cache after reading a large file
Purge disk cache (manual, helper)
2.80 GiB
39.9%
re-reading the file took 578 ms vs 84 ms from cache
Daily automatic check (can be turned off) or "Check now".
The download is verified against its published SHA-256, then installed in place.
Windows: a silent setup upgrade, or an exe swap for portable copies.
macOS: a bundle swap.
The app restarts with your settings.
Releases also publish raw MemManager-<ver>-windows-<arch>.exe assets for portable updates.
Upgrading from 0.1.0
0.1.0 has no updater, so install this version once by hand: run the new setup, or drag the new
app over the old one. Your settings are kept. Later versions update from inside the app.
Known limitations
Unsigned builds:
Windows SmartScreen may warn.
On macOS, right-click → Open the first time.
Not yet hand-tested: the dashboards are verified headless in CI only.
macOS cache-hit counter: macOS has no system-wide cache-hit counter, so the "data back from
cache" line is Windows-only.
Install
Windows 10/11 (x64 or ARM64): run MemManager-0.2.0-windows-setup.exe. Keep "Start at login with memory-cleaning privileges" checked so MemManager can manage memory without asking for admin each time. Uninstall from Settings → Apps.
macOS 13+: open MemManager-0.2.0-macos-universal.dmg and drag MemManager to Applications. On first launch, right-click → Open (the build is not notarized yet).
Portable Windows zips and a macOS zip are attached as well.
Measured memory reduction
Each row creates real bloat on a GitHub-hosted runner, runs the action through the same code path the app uses, and measures memory freed plus the cost afterwards. Numbers vary by machine and workload.
Windows (CI runner, 15.99 GiB RAM, build 26100)
Bloat
Action
Freed
% of RAM
Cost afterwards
Idle apps holding private memory
Selective trim of idle apps (tier 4)
2.02 GiB
12.6%
re-touch 514 ms vs 16 ms warm; 1 hard page faults
Modified (dirty) pages from trimmed apps
Flush modified list (tier 4)
1.57 GiB
9.8%
pagefile writes; pages become reusable cache
File cache after reading a large file
Clear low-priority cache (tier 2)
0.00 GiB
0.0%
none: only cache Windows was about to discard
File cache after reading a large file
Clear standby cache (tier 3)
5.81 GiB
36.3%
re-reading the file took 4915 ms vs 265 ms from cache
All of the above at once
Deep clean (manual)
5.21 GiB
32.6%
624111 hard faults in the next 10 s (baseline 454); apps re-touch 19509 ms
Identical pages across the system
Combine pages (tier 5)
0.02 GiB
0.1%
CPU scan only
macOS (CI runner, 7.00 GiB RAM)
Bloat
Action
Freed
% of RAM
Cost afterwards
Volatile purgeable memory (app caches)
Nudge (helper)
0.70 GiB
10.0%
apps regenerate purged caches on next use
App holding a cache, cooperative
Nudge (helper)
0.70 GiB
10.0%
the app rebuilds its cache when needed
File cache after reading a large file
Purge disk cache (manual, helper)
2.85 GiB
40.8%
re-reading the file took 563 ms vs 86 ms from cache
Lowers the memory priority of idle background apps.
Clears the priority-0 standby cache.
Clears the full standby cache, only when Windows is already repurposing it.
Selectively trims idle, silent, background apps, and flushes the modified list.
In the Aggressive profile: combines pages and caps the file cache.
Regret tracking: every action records the disk page-ins it causes, and actions that hurt
are backed off automatically.
Game mode: keeps standby clear while a full-screen app runs and never trims during games.
Leak and runaway alerts: robust trend statistics (Theil–Sen + Mann–Kendall) with a
time-until-trouble estimate, a commit-limit warning, and kernel pool-tag attribution for
driver leaks.
Tray icon: ring, bar or number styles; system, traffic-light, mono, colour-safe or custom
colours.
Fluent flyout: dashboard and settings, rounded corners, optional acrylic background.
Installer: one setup exe for x64 and ARM64. It optionally registers a logon task so
MemManager starts with cleaning privileges without a UAC prompt each time, and uninstalls
cleanly.
Footprint: about 2.6–2.9 MB private bytes when idle; about 650 KB executable.
macOS (menu bar app, Swift + AppKit)
Event-driven monitoring of kernel memory pressure, with Activity-Monitor-consistent
figures: App, Wired, Compressed, Cached, Swap.
Per-app leak alerts, summing every helper process, with Quit / Relaunch / Ignore actions.
Also a swap-thrash warning and idle-heavy-app suggestions with an opt-in auto-quit list.
Optional privileged helper:
Nudge asks apps to release caches and empties purgeable memory. It is latch-safe: the
real pressure level is always restored.
Manual disk-cache purge.
Menu bar icon: ring, bar, percent, graph or dot, with colour presets.
Install: drag-to-Applications DMG.
Footprint: about 6 MB.
Known limitations
Unsigned builds:
Windows SmartScreen may warn.
On macOS, right-click → Open the first time.
The macOS helper (Nudge/Purge) can only be installed from a Developer-ID-signed build.
Not yet hand-tested: the UIs are built and CI-verified headless only.
Accessibility: Windows screen-reader (UI Automation) support for the flyout is pending.
Install
Windows 10/11 (x64 or ARM64): run MemManager-0.1.0-windows-setup.exe. Keep "Start at login with memory-cleaning privileges" checked so MemManager can manage memory without asking for admin each time. Uninstall from Settings → Apps.
macOS 13+: open MemManager-0.1.0-macos-universal.dmg and drag MemManager to Applications. On first launch, right-click → Open (the build is not notarized yet).
Portable Windows zips and a macOS zip are attached as well.
Measured memory reduction
Each row creates real bloat on a GitHub-hosted runner, runs the action through the same code path the app uses, and measures memory freed plus the cost afterwards. Numbers vary by machine and workload.
Windows (CI runner, 15.99 GiB RAM, build 26100)
Bloat
Action
Freed
% of RAM
Cost afterwards
Idle apps holding private memory
Selective trim of idle apps (tier 4)
1.99 GiB
12.4%
re-touch 454 ms vs 8 ms warm; 434 hard page faults
Modified (dirty) pages from trimmed apps
Flush modified list (tier 4)
1.52 GiB
9.5%
pagefile writes; pages become reusable cache
File cache after reading a large file
Clear low-priority cache (tier 2)
0.00 GiB
0.0%
none: only cache Windows was about to discard
File cache after reading a large file
Clear standby cache (tier 3)
7.98 GiB
49.9%
re-reading the file took 6069 ms vs 232 ms from cache
All of the above at once
Deep clean (manual)
5.07 GiB
31.7%
975881 hard faults in the next 10 s (baseline 1788); apps re-touch 40483 ms
Identical pages across the system
Combine pages (tier 5)
0.03 GiB
0.2%
CPU scan only
macOS (CI runner, 7.00 GiB RAM)
Bloat
Action
Freed
% of RAM
Cost afterwards
Volatile purgeable memory (app caches)
Nudge (helper)
0.69 GiB
9.9%
apps regenerate purged caches on next use
App holding a cache, cooperative
Nudge (helper)
0.70 GiB
10.0%
the app rebuilds its cache when needed
File cache after reading a large file
Purge disk cache (manual, helper)
2.68 GiB
38.3%
re-reading the file took 677 ms vs 189 ms from cache
