Skip to content

FAQ

Answers about Yap releases, platform support, privacy, model selection, shortcuts, local runtimes, and unreleased features.

What is Yap?

A free, MIT-licensed desktop dictation app. Record speech, transcribe locally with whisper.cpp or through OpenAI, and copy or auto-paste the result into another application.

Which version do these docs cover?

The latest published stable release is 0.5.0. It includes the Models Library, text-file saving, runtime fixes, and signed/notarized macOS packages with Homebrew installation. Screenshots show the release React UI with sample data. See release status.

Which platforms can run the downloads?

The macOS DMG is Apple Silicon arm64. Windows releases are x64, with installer and portable ZIP options. Linux provides x64 AppImage and .deb variants for WebKitGTK 4.0 and 4.1 environments. Read installation before picking a package. The ARM Mac app does not run on Intel Macs through Rosetta.

Does it work offline?

Local transcription does, after the model has downloaded and the runtime is ready. Release packages include the runtime. OpenAI mode requires internet and a separately billed API account.

Is my data private?

Local transcription processes audio on your device. OpenAI mode sends audio to OpenAI. History, recordings, and settings remain local files; Yap does not encrypt them. The OpenAI key is plain JSON in the config file. Review configuration and OpenAI’s API data controls.

Does Clear History remove the audio?

No. It clears history records, but does not clean the recordings directory. Deleting an individual item attempts to remove its associated audio. See History.

Can I save transcripts as text files?

Yes. Choose an output folder in Settings and enable Save transcriptions as .txt. It is off by default and saves future successful transcriptions from either provider, not existing history. These files remain when you delete or clear history. See Settings for saving and error recovery.

Which model should I use?

Start with base.en for English, or base/small for another language. Compare quality and processing time on your own machine. Larger models cost more memory and time; they are not a correctness guarantee. Version 0.5.0 includes the multilingual automatic language-selection fix.

Is cloud always faster or more accurate?

No. Results depend on your recording, model, hardware, network, and service. Local mode has no usage fee; OpenAI charges according to its current pricing.

Can I change the hotkey?

Yes. Settings captures a single recording key and a separate cancel key. Defaults are Right Option on Mac, Right Alt on Windows/Linux, and Escape to cancel. Multi-key combinations and the old documented double-tap preset are not the current interface. See keyboard shortcuts.

Does it use a GPU?

Version 0.5.0 has Metal-first inference on Apple Silicon with CPU fallback. Do not assume Windows/Linux packages include GPU acceleration.

Are Obsidian integration and AirPods media mode available?

They are not listed as 0.5.0 features. Check their links on the release-status page for current progress. Normal microphone selection does not imply the proposed AirPods mode exists.

Can I import audio files or issue voice commands?

The current app’s workflow records speech and transcribes it. It has no user-facing audio-file import or voice-command system. It does not execute what you dictate.

Where do I get help?

Use troubleshooting first, then GitHub issues with your version, platform, reproduction steps, and relevant logs.