Using Waffle

Set up dictation, choose a model, and configure vocabulary and writing styles. Processing stays on your Mac.

Start dictating

Waffle requires macOS 27 or later and an Apple silicon Mac. A public download is coming soon; this guide covers setup once you have the app.

  1. Open Waffle and choose a model. In Models, use Apple Dictation and install its language assets if prompted, or download a Parakeet model.
  2. Allow microphone access. Waffle asks when you start a recording. Choose your preferred input in Settings → Microphone.
  3. Enable Accessibility for automatic paste. Open Settings → Permissions in Waffle. Without access, Waffle copies your transcript so you can paste it yourself.
  4. Put your cursor where you want to write. Press Control + Option + Space. Wait for Listening, then speak.
  5. Finish your thought. Press the shortcut again or Return. Waffle transcribes and delivers your words to the app where you started.

Prefer to hold a key? Hold the dictation shortcut for at least 0.3 seconds, speak, and release to finish. Press Escape to cancel.

Hot keys

These are the defaults. Change the start, finish, and cancel keys in Settings → Hot Keys.

ActionDefault shortcut
Start or finish dictationControl + Option + Space
Finish a recordingReturn
Cancel a recordingEscape
Insert a reviewed transcriptCommand + Return
Search HistoryCommand + F

You can also give a writing style its own hot key in the style editor, or select it from Start Dictation in Style… in the menu bar.

Speech models

Apple Dictation

Shows live text and supports vocabulary hints. Choose your language in Settings, then install any required language assets in Models. To see words in the floating capsule, enable Settings → Recording Overlay → Show live transcript in overlay.

Parakeet

The Multilingual model supports 25 European languages, including English. The English model focuses on English-only dictation. Both transcribe after you finish recording, using downloaded models on your Mac.

For extra help with custom vocabulary, install the Vocabulary Add-on in Models and enable Boost vocabulary recognition. The add-on checks English audio and can help with your saved terms; recognition is not guaranteed.

Want a quicker start? Settings → Performance → Keep speech model ready retains models between recordings. It uses more memory, but the microphone still stops when the recording ends.

Writing styles and cleanup

In Writing Styles, choose Literal to preserve your wording, Natural for light cleanup, or Tight for more concise text. Fine-tune fillers, repetitions, and grammar separately from tone.

Edit Personal, Work, or Email, or add a style of your own. Set its tone, optional cleanup override, and instructions. Under Use automatically in, assign apps or website domains and enable Apply styles automatically.

Website rules include subdomains and take precedence over app rules. If the browser doesn’t expose its URL, Waffle uses the app’s style. A style selected manually or by its hot key overrides automatic matching for that recording.

Try changes in Playground before saving them. Record a sample, compare the original and result, and use Apply Settings Again to test different cleanup choices on the same transcript. Playground doesn’t paste automatically or save to History or Stats.

AI cleanup and tone changes need Apple Intelligence to be available on your Mac. Vocabulary capitalization, saved corrections, and spoken shortcuts also work without it. For output without AI edits, choose Literal and turn off tone changes, formatting, and spoken-correction recognition.

Vocabulary and corrections

Open Vocabulary and add names, tools, or phrases with the capitalization you want. Press Return after each entry. You can save up to 100 terms.

Use Corrections below the vocabulary list for repeat misspellings, such as x code → Xcode or type script → TypeScript. Corrections match whole words and phrases without regard to capitalization, then insert your exact replacement.

Changes apply to your next recording, across every model. They do not rewrite the live preview or old transcripts.

Spoken shortcuts

Open Shortcuts, enter a spoken phrase and the exact text it should insert, then choose Add. For example, save “my sign off” as your name and signature. Use Option + Return for a new line in a snippet.

A shortcut spoken on its own skips AI cleanup and returns the saved text. In a longer sentence, shortcuts expand after cleanup and spelling corrections. Snippets are always text; they never execute commands.

Privacy and storage

Audio and recognition. Speech recognition runs on your Mac. Audio is held in memory for the current session; Waffle does not save or upload it. Optional cleanup uses Apple Intelligence on-device. Model installation uses the internet: Apple supplies native assets and Parakeet downloads from Hugging Face. Damaged model assets may be downloaded again.

History and Stats. Transcript history is opt-in and keeps the latest 100 results on your Mac. Stats store text-free usage records, including dates, word counts, durations, models, and destinations, independently of History. Clear each separately in the app. The latest transcript stays in memory until you quit.

Cursor context. When enabled, nearby text and your selection from supported fields help with tone and insertion. Password fields are excluded. This context stays in memory for the current dictation and is not saved in History or synced.

Optional iCloud sync. Off by default. When enabled in Settings → iCloud Sync, preferences such as vocabulary, corrections, styles, and snippets use your Apple Account’s iCloud storage. Transcripts, audio, statistics, microphone choice, model downloads, and hot-key assignments stay local.

No Waffle account or analytics. Waffle has no separate account or embedded analytics. This website has no trackers, cookies, or third-party font requests. An explicit appearance choice is stored locally in your browser.

Troubleshooting

My transcript is copied, but doesn’t appear in the other app.

Enable Accessibility for Waffle in System Settings → Privacy & Security → Accessibility. Check that Waffle’s output is set to Paste to active app. If the target field or destination changed during recording, Waffle may leave the text on your clipboard instead. Paste it manually, or copy the last transcript from the Waffle menu.

I’m not seeing live words while I speak.

Live text is available with Apple Dictation. Enable Show live transcript in overlay under Settings → Recording Overlay. Parakeet transcribes after you finish speaking, so it displays the waveform during recording.

The microphone isn’t picking anything up.

Check microphone permission in System Settings → Privacy & Security → Microphone. Then use the input test in Waffle’s Settings → Microphone. If a selected device disconnects, Waffle stops recording; choose an available input before trying again.

Cleanup or a writing style isn’t available.

Check the Apple Intelligence status in Writing Styles. Your Mac needs Apple Intelligence enabled and available for AI cleanup and tone changes. You can still dictate, apply saved corrections, and use spoken shortcuts without it.

Can I check the text before it’s pasted?

Yes. Enable Settings → Output → Review before delivery. Edit in the floating review panel, use Show original to compare, and press Command–Return to insert. Discarding a pending review does not deliver or save it.