/ tools / holy shift
It listens while you talk and drops a very loud beep the second you swear. Everything happens on your own machine — and it forgets what you said as it goes.
What it looks like
What it keeps
Short version: the number, and nothing that could reconstruct the sentence it came from.
| The audio | Held in memory about fifteen seconds at a time and thrown away as it is transcribed. Never written to disk, never uploaded. |
|---|---|
| The transcript | Exists only while a session is running, and only in memory. Stopping the session deletes it. You can switch it off entirely and still get the beep and the count. |
| The counts | A tally per word, how long each session ran, and how many seconds in each slip landed. That last one is what makes "cleanest hour" and "best streak" possible. |
| The network | Used exactly once, to fetch the speech model on first run. After that it works with the wifi off, and there is no account, no sync and no telemetry. |
The details
Seven beeps, or your own
Classic censor tone, double beep, air horn, bell, boing, buzzer, or a barely-there tick. Or point it at any .wav or .mp3 you like. Volume, length and the gap between beeps are yours.
The list is editable
Mild, strong and blasphemy tiers you can switch on and off, plus anything you want to add. Blasphemy is off by default, because "god" turns up in ordinary sentences all day.
It can be told it's wrong
Every catch offers "Not a swear". That takes the slip back off the count and teaches it never to flag that word again. "Classic", "assassin" and "shift" are already exempt.
About a second behind
Speech recognition needs a window of audio, not a syllable, so the beep lands shortly after the word rather than on top of it. It's a "you said it" buzzer, not a broadcast delay.
Installing it
Windows will warn you on first run
The installer isn't code-signed — a certificate costs a few hundred dollars a year, which is hard to
justify for something shared with friends. Click More info, then Run anyway.
You'll only see it once. It installs under your own account, so there's no admin password, and it
uninstalls from Settings → Apps.
First launch asks you to pick a speech model and downloads it — 75 MB, 150 MB or 490 MB depending on how accurate you want it. That's the only time it touches the network.