Microsoft PowerToys Preview v0.101.2323.0 lands with critical regression fixes
Following a few weeks of preview channel churn, Microsoft released PowerToys Preview v0.101.2323.0 on August 21, 2026. This is a pure maintenance drop. No new modules. No experimental toggles. Just four merged PRs that patch regressions threatening to undo your actual workflow.
If you run the preview channel, you have probably already felt some of the rough edges in the 0.101 series. The headline fix here addresses a fairly nasty setting-persistence bug. Check out PR #50018, which closes issue #48907. If you have ever triggered an update check only to watch your enabled PowerToys quietly disable themselves, this is for you. The runner was treating the update-check IPC payload as a full general-settings dump. If the UI loaded fallback defaults during that handshake, it would persist those defaults over your actual configuration. Not exactly a great look.
The fix strips the IPC message down to just the action itself. The runner no longer sees an update trigger as a settings overwrite. It preserves the prerelease toggle behavior because that already sends normal general-settings IPC before kicking off the actual update. Brought to you by Niels Laute, with Copilot handling about half the code review now.
What actually broke (and how it got fixed)
The second regression lives in Shortcut Guide. PR #50046 closes issue #50030 and restores excluded-app filtering for both regular hotkeys and Windows-key-hold activations. A previous refactor switched the module to a persistent background process. That meant the exclusion gate only ran once at startup. Newly saved rules died on arrival. The fix forces the listener to evaluate the current foreground app before every overlay activation, which keeps your workflow exactly as you configured it.
On top of that, the automated issue triage bot got a reality check. PR #49999 patches two workflow regressions. The CI was actively mutating version labels on bug reports, and its action-detection logic only recognized the verb "press." If you wrote "make" or "hold" in your reproduction steps, the bot classified your report as insufficient and slapped a feedback tag on it. The script now handles natural-language reproduction steps. A quick run of the 50 unit tests confirms the evaluator now marks previously flagged issues as sufficient.
The fourth merge is lighter fare. PR #49962 just adds an attribution link on the Window Hopper settings page to credit Wzhudev's work on AltBackTick. Small touch, but honestly, the community repo has 138k stars and nearly 8.5k forks at this point. Acknowledging the people actually keeping the lights on is about as important as any feature drop. Copilot handled the code reviews across all four PRs. Some people bristle at AI-assisted development, but it is hard to argue with a workflow that catches missing edge cases faster than manual review. The code itself is still human-written. The review just got a second pair of eyes that never gets tired.
Not every fix is a fire alarm. At this point, the preview channel runs on a pretty tight feedback loop. You drop a build, the triage bot gets confused by natural language, someone patches it, and the cycle continues. It is not glamorous, but it is exactly how stable software actually gets built.
Grab it yourself
Installers are available for both x64 and ARM64 architectures, with separate per-user and machine-wide builds. The per-user ARM64 installer sits at roughly 288 MB, while the machine-wide x64 build pushes closer to 297 MB. If you plan on debugging the source yourself, the symbols packages run roughly 472 MB for ARM64 and 483 MB for x64. You can grab the exact SHA-256 hashes and download links directly from the GitHub release page.
Head here to download the preview installer. Keep in mind that this is a prerelease track intended for power users who want to test fixes before they land in a stable release. If your setup relies on specific module configurations, you might want to run a settings backup before swapping versions. Either way, the regression fixes here should quiet a lot of the noise in the 0.101 series.
