wispr-fox v3.1.0-nightly.1

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v3.1.0-nightly.1 — Noise reduction for fan-heavy laptop mics

Built-in laptop mics pick up the machine's own fans — a low whirring bed under
everything you say that makes the transcriber mishear words. This nightly adds
an opt-in noise-reduction stage that scrubs that noise out of the audio sent
for transcription, tuned on real recordings from the affected machines.

Noise reduction — Off / On / Aggressive

New in Settings → Dictation → Noise reduction:

• Off (default) — nothing changes; audio goes out exactly as recorded.
• On — a rumble filter that removes the deep fan hum below ~90 Hz. It cuts
  the loudest part of the fan signature and by design cannot touch speech —
  safe to leave on permanently.
• Aggressive — the rumble filter plus an AI speech denoiser (RNNoise, the
  same family of tech Discord and OBS use). On the test recordings this took
  the noise floor down by 20–45 dB and roughly doubled-to-tripled the
  signal-to-noise ratio. If a transcript ever comes out worse, just step back
  down to On.

Built so it can't hurt you

• Your saved recordings are untouched. The cleanup happens on a temporary
  copy that only the transcription provider sees. History always keeps and
  plays the original audio, so you can compare and diagnose after the fact.
• No noticeable wait. Processing runs locally at roughly 130–600× realtime:
  a 10-second dictation costs ~15–80 ms, a full minute well under half a
  second — background noise compared to the transcription request itself.
• The avatar tells you what's happening. The floater shows a quick
  "clearing noise…" beat for exactly as long as the denoiser runs, and each
  recording's flight recorder logs the precise time it took (visible in the
  per-recording timeline).
• Never costs a dictation. If the denoiser hits any problem, the pipeline
  quietly falls back to the raw recording and carries on.

Notes

• Applies to hotkey dictations. Uploaded audio files are transcribed as-is.
• Desktop only for now (Windows + macOS + Linux builds all include it).

---

Built from commit 742e116.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v3.0.0

wispr-fox v3.0.0

The big one. Since v2.1.0, wispr-fox grew from a single-machine dictation
tool into something that follows you across every device you use it on — and
this build makes that whole line stable.

Your dictation, on every device

Sign in (optional — Google or email/password) and your transcripts and API
keys travel with you across the desktop app, the phone app, and the web app.
Talk on your laptop, and it's in your history on your phone. Set your keys
once; every signed-in device has them.

Your audio never leaves the machine it was recorded on. Only the text
syncs. Signed out, nothing changes — the app works exactly as it always did,
no account, no cloud, bring-your-own-key.

Transcribe audio files, not just your voice

Drag an audio file onto your history — or use the Upload button — and
wispr-fox runs it through the same transcribe → clean → draft pipeline as
live dictation. Voice memos, call recordings, anything your machine can play.
Pick the provider and cleanup style per batch; uploaded items are badged so
you can tell them apart.

Delete now means "mine"

Deleting a transcript used to open a little matrix of choices — this device
or everywhere, the audio or the text. That was more decision than anyone
wanted, and across synced devices it had become risky: a "delete all" could
reach out and wipe history you'd recorded somewhere else.

The rule is one line now: you can delete what this device recorded, and
nothing else. Deleting a transcript takes its recording with it.
Transcripts that came from your phone or the web app show in your history,
but there's no delete button on them here — they belong to those devices.
"Delete all" clears only what this desktop made.

Purge — the clean-slate button

That rule left one gap: a transcript recorded on a device you no longer have
— an old phone, a reinstalled machine — could never be deleted, because no
remaining device "owns" it.

Settings → Account → Purge closes it. Purge wipes every transcript on
your account, on every device, including ones you no longer have. Each of
your other devices clears its own copy the next time it syncs. It asks twice
— a press-and-hold, then a confirm — because it reaches further than anything
else in the app and can't be undone. Try it deliberately the first time.

Also in this build

• Settings sections are reordered into one consistent shape shared with the
  phone and web apps — Account sits near the bottom, where account things
  live, with Purge at its foot.
• A quiet fix: deletes made on the desktop weren't always reaching your other
  devices. They are now.
• Synced-history polish — action buttons stay aligned on rows that came from
  other devices, and each row shows a small chip for where it was recorded.

---

wispr-fox is three apps sharing one backend: this desktop app, the
Android app, and the
web app. They all share the delete and
purge behavior described above.

---

Built from commit 113b401.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v3.0.0-nightly.3

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v3.0.0-nightly.3

More work on the accounts + cross-device line — this build is about delete
that behaves the same everywhere, and a way to wipe the slate clean.

Delete now means "mine"

Deleting a transcript used to offer a little matrix of choices — this device
or everywhere, the audio or the text. That was more decision than anyone
wanted, and on the web it had become genuinely dangerous: a "delete all" could
reach across and wipe history you'd recorded on another machine.

So the rule is one line now: you can delete what this device recorded, and
nothing else. Deleting a transcript takes its recording with it. Transcripts
that came from your phone or the web app show up in your history, but there's
no delete button on them here — they belong to those devices. "Delete all"
clears only what this desktop made.

Purge — the clean-slate button

There was a gap in that rule: a transcript recorded on a device you no longer
have — an old phone, a reinstalled machine — could never be deleted by anyone,
because no remaining device "owns" it.

Settings → Account → Purge closes that gap. It wipes every transcript on
your account, on every device, including ones you no longer have. Each of your
other devices clears its own copy the next time it syncs. It asks twice — a
press-and-hold, then a confirm — because it reaches further than anything else
in the app and can't be undone.

Also in this build

• A quiet but real fix: deletes made on the desktop weren't always reaching
  your other devices. They are now.
• Settings sections are reordered into a consistent shape shared with the web
  and phone apps — Account sits near the bottom, where account things live.
• The synced-history polish from the last couple of pushes (aligned action
  buttons, source chips beside the version tabs) rides along here.

---

A note on testing: the account and sync paths in this build are checked by the
compiler and the type system, but this is a nightly — the delete and purge
round-trips haven't been exercised against a live account yet. Purge especially
is worth a careful first try, since it's designed to reach every device.

---

Built from commit be3e5a1.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v3.0.0-nightly.2

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v3.0.0-nightly.2 — Sync actually syncs now

A focused fix for the sync that shipped in nightly.1.

"Sync paused — will retry" is fixed

If you signed in on nightly.1 and saw "Sync paused — will retry" sitting
under your account (even right after a clean sign-in), that's fixed. Sign-in
itself was fine — the very first step of the sync, registering your device with
the cloud, was being rejected, so the whole cycle stopped and quietly kept
retrying against the same wall.

Sign in again (or hit Sync now in Settings → Account) and it should move to
Syncing… and then settle on a normal "last synced" time. Your existing
transcripts push up on that first successful sync, and anything from your other
devices pulls down.

API keys now sync from desktop too

The same underlying issue was quietly blocking your API keys from syncing up
from the desktop, so set up a provider here and it's now available on your web
and mobile apps as well — no re-pasting.

Nothing else changed — local-only mode, dictation, and everything from nightly.1
behave exactly as before.

---

Nightly build — v3.0.0 line. Signed out, wispr-fox works fully local as always.

---

Built from commit 22cd8f0.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v3.0.0-nightly.1

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v3.0.0 — Accounts + cross-device sync

The big one. wispr-fox can now carry your transcripts with you — sign in and
they follow you across desktop, web and mobile. Sign-in is completely optional:
signed out, wispr-fox works exactly as it always has, fully local, no account.

Sign in — Google or email

There's a new Account section in Settings, and a new (skippable) step at the
end of onboarding.

• Continue with Google opens your browser, you approve once, and you're
  back — no passwords to manage.
• Prefer email? Create an account or sign in with an email and password
  right in the app.
• Not ready? Continue without an account is right there, front and centre.
  Nothing changes for you.

Your transcripts, everywhere

Once you're signed in, your transcripts sync in the background across every
device you use wispr-fox on:

• Everything you've already recorded gets pushed up the first time you sign in,
  and new dictations sync as you make them.
• Transcripts made on other devices show up in your History here, each with a
  small Desktop / Web / Mobile badge so you can tell where it came from, and
  your device's name on hover.
• Sync now and a "last synced" time live in Settings → Account, so you can
  always see what's happening. Sync is quiet and best-effort — if the network
  hiccups it just tries again, never interrupts you, never slows down dictation.

Your API keys come along too

Set up Deepgram (or Groq, OpenAI, ElevenLabs, Gemini) on your laptop and it's
ready on your other devices — your provider keys sync securely so you don't have
to paste them in again on every machine.

Voice recordings stay put — always

Only text syncs. Your audio never leaves the device that recorded it. This
is deliberate: it keeps the whole thing free and keeps your voice private.
Transcripts synced from another device show up as text you can read, copy and
search — there's just no audio to play back, because it was never uploaded.

A calmer, clearer delete

Deleting is reworked so you're always in control of exactly what goes:

• The delete control is now press-and-hold — hold it for a moment (a little
  fill sweeps to confirm) and a dialog opens. A quick tap just reminds you to
  hold, so nothing disappears by accident.
• The dialog lets you choose what to remove — voice files, transcripts, or
  both — and where: this device only, or everywhere (when you're
  signed in). It spells out the consequence in plain words before you confirm.
• Everywhere removes the transcript from all your devices. This device
  only clears it here and won't let it sync back. Deleting just the voice
  files frees up disk while keeping the text.

A note on privacy

Local-only mode is unchanged, byte for byte — if you never sign in, wispr-fox
behaves exactly like it did before, storing everything on your machine and
talking only to the transcription provider you chose with your own key. Signing
in adds transcript + key sync on top; your audio is never part of it.

---

Nightly build — the first of the v3.0.0 line. If this build shows "Sync not
configured", the sync backend isn't wired into this particular build yet;
everything else works as normal and it behaves as signed-out.

---

Built from commit b6a2298.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v2.2.0-nightly.1

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v2.2.0-nightly.1

Now it transcribes files too — not just your microphone.

Upload audio to transcribe

You've already recorded something — a voice memo on your phone, a meeting on
your laptop — and you just want the text. Drop it in.

• Drag any audio file onto the window and wispr-fox stages it and opens the
  upload panel. Or hit the new Upload button on the History page and pick
  one (or several) files from your computer.
• Choose your engine right there. The upload panel lets you pick which
  transcription provider and model to use for this batch — independent of your
  usual dictation default — and tick Clean up and/or Draft to run the
  same polishing you get on a live dictation.
• It all lands in your history, in line with everything else, each file
  marked with an Uploaded badge. Play the audio back, read the Raw /
  Cleaned / Drafted tabs, copy the text, open the (i) panel — exactly like a
  recording you made with the hotkey.

Supported formats: wav, mp3, m4a, aac, ogg, opus, flac, webm — so notes
from a Samsung/iPhone recorder, Windows Voice Recorder, or almost anything else
just work. Long files are fine; for very large files, Deepgram handles them
most smoothly.

Your files never leave your machine except to go straight to the transcription
provider you chose with your own key — same bring-your-own-key model as the
rest of the app.

---

Nightly build. Built on the v2.1.0 stable line.

---

Built from commit 9ddeb4c.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v2.1.0

wispr-fox v2.1.0

The dictation gets quieter, the history gets smarter, and your laptop can
finally sleep. This is the stable release of everything the v2.1.0 nightly
cycle has been building since v2.0.0 — pulled together, tested, and promoted
to Latest.

Your laptop can sleep again

The headline fix. A background audio helper used to hold your sound device
open for the app's entire lifetime — and Windows treats any app that's holding
the audio device as "audio is playing," so it refused to ever go to sleep.
wispr-fox now opens that device only for the split second it plays a start/stop
cue and releases it after a few seconds idle. Nothing about the cues changes;
your machine just sleeps normally again.

A roster of characters — including pixel pets

• Pixel pets. Eight animated sprite-buddies now live in the avatar picker
  alongside the watercolor fox — they idle, listen, and celebrate as you
  dictate. Pick your favourite from the sidebar (six featured, plus a "More"
  tile for the rest).
• Minimal skins. Prefer no character at all? The wave skin is a small
  translucent pill with a live waveform that dances to your voice, and the
  siri skin is a glowing orb. Both stay out of the way.
• Show/hide, your call. A three-way avatar toggle: always visible, only
  while you're dictating, or fully hidden.

History that names itself

• Auto-titles. Each recording now gets a short one-line name written for
  it, so your history reads like a table of contents instead of a wall of
  timestamps.
• A built-in flight recorder. Open the (i) panel on any recording to see
  exactly how long transcription and cleanup took, plus a step-by-step
  timeline. A slow result now explains itself.

Your mic, diagnosed

Two classes of "why is my transcript wrong" now get caught and explained
instead of silently pasting a broken result:

• Mic dropped mid-recording. If your microphone dies partway through
  (a known Bluetooth-headset quirk), the app compares how much audio it
  actually captured against how long you spoke and tells you the transcript is
  cut short — rather than pretending a half-recording succeeded.
• Mic slow to wake up. On some Windows machines the mic takes several
  seconds to start after you press the key, quietly eating your first
  sentence. The app now measures this "wake-up time" on every recording, and
  the onboarding demo doubles as a mic health-check — with the exact Windows
  settings to fix it when it's slow (turn off the mic's audio enhancements and
  "exclusive control").

A first-run that actually walks you through it

Onboarding was rebuilt from the ground up. It fills the window cleanly at any
size (no more frozen band with white margins), acts out a real dictation on the
welcome screen so you can see what the app does, and walks you through picking
a transcription engine and pasting a key with a clear "I already have a key"
vs. "help me get one" fork. Setup takes about five minutes and the app is, as
always, just the interface — you bring your own key and it stays on your
machine.

---

Bring-your-own AI key. No subscription, no account, no telemetry. Windows and
macOS (Apple Silicon); the macOS build is unsigned — right-click → Open on
first launch.

---

Built from commit 43e7126.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v2.1.0-nightly.9

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v2.1.0-nightly.9

Your laptop can sleep again — and onboarding got a proper do-over.

Fixed: wispr-fox was quietly keeping your PC awake

If your laptop stopped going to sleep on its own, this was us. The little
"dop" sound that plays when you start and stop a recording kept its audio
channel open forever after the first use — and Windows treats any open
audio stream as "someone is listening to music, don't sleep." One dictation
after launch and your machine would stay awake until you quit the app.

The cue sound now opens its audio channel only when it actually plays and
releases it half a minute later. Nothing changes about how cues sound or
feel — your power settings just work again.

(If you want to double-check on your own machine: run powercfg /requests
in an elevated terminal. Before this build, wispr-fox showed up under
"An audio stream is currently in use." Now it doesn't.)

Onboarding, rebuilt

The first-run experience is a clean three-step story again:

• See it before you set it up. The welcome screen acts out a real
  dictation: a hotkey cap gets pressed, one of the pixel buddies listens,
  transcribes, and the words type themselves into a little box. The buddy
  changes every loop, so you meet the avatar roster before you ever open
  Settings.
• A straighter path to your key. Pick Deepgram (recommended — $200
  signup credit that lasts years of daily use) or Groq (free forever),
  then tell us where you stand: "I already have a key" drops you at a
  paste box; "Help me get one" walks you through signup and the keys
  page link by link. The optional cleanup brain (Gemini, free) appears
  once your engine key is verified — and on the Groq path it's marked
  done automatically, because one Groq key does both jobs.
• Fits your window. No more fixed-width band with odd gutters, and no
  scrolling at the default window size — every step fits on one screen.

Nightly build. The demo step is unchanged: press your dictation key on the
last screen and watch your words land in the box.

---

Built from commit 8d10761.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v2.1.0-nightly.10

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v2.1.0-nightly.10

Your mic gets a health-check — starting with the onboarding demo.

The "it ate my first sentence" detector

On some Windows machines the microphone takes 3–8 seconds to actually wake
up after you press the dictation key. You start talking immediately; the
audio only starts flowing seconds later; your first words are simply gone.
Nothing in the app could see this before — the recording timer starts late
too, so everything looked fine.

The app now measures mic wake-up time on every recording: the gap
between your key going down and the first real audio arriving.

• In the onboarding demo, the try-it step now doubles as a mic
  health-check. After your first recording you'll see a verdict: a green
  "instant handover" chip when things are healthy, or a clear warning with
  the exact settings to fix when they're not — on Windows that's almost
  always the mic's audio enhancements and "allow applications to take
  exclusive control" toggles. Change them, press the key again, watch the
  number drop.
• In normal use, if your mic is pathologically slow to wake (over
  ~2.5s), the floater tells you once per session — with the same fix —
  instead of letting you lose the start of every dictation forever.
• In the (i) panel, every recording's timeline now includes its
  "mic wake-up" number, alongside the recorded / captured / provider
  triangulation from nightly.8. Between the two, a truncated transcript
  now tells you where the audio went missing: the head (slow wake-up —
  fix your mic settings), the middle/tail (mic dropped — often Bluetooth),
  or in transit (provider saw less than was captured).

Nightly build. If a recording ever comes back clipped, open its (i) —
the numbers will point at the culprit.

---

Built from commit 41d7f89.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.

wispr-fox v2.1.0-nightly.8

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v2.1.0-nightly.8

Catching the "3 minutes in, half a transcript out" bug.

The likely culprit behind a truncated transcript

If you've ever recorded for a few minutes and got back only the first
30 seconds — this build is about that.

The recording timer measures how long you held the key. It does not
prove that much audio actually reached the file. If the microphone stream
hiccups partway through — Windows switching the device, a Bluetooth headset
changing modes, an audio "enhancement" kicking in — the app used to keep the
timer running while quietly capturing nothing. You'd get a card that says
"3m 0s" wrapped around 30 seconds of audio, and a transcript cut short with
no explanation. Retrying just re-read the same short file, so it never helped.

Now the app measures the audio it actually captured and compares it to the
timer. When they don't match, you get told plainly:

Mic dropped mid-recording. Only ~30s of your 3m 0s recording was
captured (~150s lost), so this transcript is cut short. Re-record to get
the rest — retrying won't recover audio that wasn't captured.

No more silent partial results dressed up as success.

The (i) panel now triangulates

Open any card's (i) and the timeline shows three numbers that pinpoint where
audio went missing, if it did:

• recorded — how long the timer ran.
• captured — how much audio actually made it into the file.
• provider processed — how many seconds the transcription service saw.

If recorded and captured disagree, your mic dropped (a re-record fixes it).
If they agree but processed is short, the problem is upload/transport, not
your mic. Either way you can now see which, instead of guessing.

Cleanup no longer gives up early

The polish/Draft step had a tight time limit that a longer dictation could
blow past, dropping you back to the raw transcript with a "cleanup timed out"
note even when nothing was really wrong. That limit is now generous enough to
finish a long draft.

Nightly build. If a transcription comes back short, open its (i) — the
recorded-vs-captured numbers will tell us exactly what happened.

---

Built from commit b32462b.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.