Something not working as expected? Start here — it covers the most common issues. If you don't see your case, email us at support@airgapscribe.com.
Only from the Microsoft Store: search for "AirGapScribe" or use the "Install free" button on this site. We don't publish installers on the web or anywhere else.
The Store installs the app and delivers its updates: there's no installer to download or run.
If someone offers you an AirGapScribe installer outside the Microsoft Store, don't run it and contact us.
Quick 1-minute diagnosis: right-click the tray icon → Tools → System test. Record 5 seconds while speaking and check the result.
The most common causes are recording for less than 5 seconds, or a microphone with very low volume.
If it still doesn't work: tray → Report a problem, which automatically attaches the technical diagnosis.
The base model (Whisper Small) ships with the installer: there's nothing to download to start transcribing. This issue only comes up if you try to switch to a bigger model (Medium or Large) from the tray.
Your network is probably blocking that download. It uses a standard CDN, usually allowed — try again from the dialog.
If it's still blocked and you have the Pro edition: tray → Tools → Import local model lets you load a model downloaded on another network.
The license key arrives by email after purchase (check spam if you don't see it).
Tray → 💎 Activate Pro / License → paste the full key → Activate.
If you switch computers, you can remove the license from the old one with Deactivate license, in that same window.
If the error persists, contact us with your machine ID ("Copy" button next to the ID in the license window) and the exact error text.
The echo filter compares microphone energy against system audio energy. With loud speakers, some echo can get through.
The most effective fix is using headphones.
Alternative: lower the speaker volume, or switch capture mode to "Microphone only" if you don't need to transcribe other participants.
All your recorded meetings show up in Your meetings, the Meeting assistant's home screen, grouped by today, yesterday, this week, this month and older.
To record, press Record meeting at the top of "Your meetings"; while recording, the button turns into Stop with the elapsed time. You can also start and stop from the AirGapScribe icon in the system tray (▶ Start recording) or with the keyboard shortcut. The button needs AirGapScribe running in the tray. When you're done, your meeting shows up in the list.
Click a meeting to open its own page: the title renames with a click, you can assign it a Project, and an action bar (Minutes, Summary, Tasks, Email, Ask) works only with that meeting.
You can also Download transcript as a file, or search your meetings by what was said, by title or by date.
On the free (Demo) version you can also open "Your meetings" and listen to each meeting; the AI action bar, fixing and reprocessing are part of Pro. Each transcript is also saved as a .txt file in your Documents\AirGapScribe folder.
Yes. Every meeting has its own player, on the free version too: use Play/Pause and turn on Follow playback so the text scrolls on its own.
To correct it: double-click a segment (Double-click to edit) to fix the text, or click a speaker's name to rename them across all their turns at once.
If a stretch came out wrong, select it and press Reprocess (or Reprocess this part inside an untranscribed gap) so the app transcribes it again on your computer. The result shows up below the original and replaces nothing until you choose Use the new version.
Listening is free. Correcting text, renaming a speaker, and reprocessing a stretch are part of Pro.
No. The assistant's AI model (1.1 GB) ships with the installer: there's nothing extra to download to start using it. The meeting assistant is part of Pro.
If you have an NVIDIA GPU, you can turn on optional acceleration from Turn on acceleration in the agent settings: it's a one-time download and works offline once installed. Measured on an RTX 3060 with the current AI model, on a short meeting: the minutes took about 78 seconds on CPU and about 6 seconds on GPU.
Without that acceleration, the assistant still runs on CPU — it's just slower.