User Guide

1. Dictation basics

Select a supported text field, tap ⌃⌥ (Control + Option), speak, then tap it again. Saywave attempts to insert the result at the active cursor. Secure fields and some custom app controls may reject automated insertion.

Prefer holding the key instead of tapping it? Open the menu-bar icon and choose Switch to Push-to-Talk — then Saywave records only while you hold ⌃⌥ and stops the instant you release it.

The dictation indicator

While you dictate, a small floating capsule appears at the bottom of your screen: a live level meter confirms Saywave is hearing you, a text preview streams in as you speak, and the capsule reports honestly when something is off — a silent microphone or a password field that can't accept pasted text. Press Esc or click the capsule to cancel a recording — the capsule closes right away. You can also drag it anywhere on screen, and it remembers its spot. Turn the indicator off entirely in Settings → Shortcut.

2. Choosing a model & language

The menu bar's Model submenu lists every speech engine Saywave supports — Whisper, NVIDIA Parakeet, and (on macOS 26+) Apple's own on-device engine — grouped by size and expected speed. Parakeet v3 is the default model and its upstream model supports 25 languages. Accuracy and speed vary by language, hardware, audio quality, recording length, and model. The live text preview shown while you dictate runs on the same model as your final result, so what you see forming is what gets pasted.

Downloaded models can remain loaded in memory while resources allow, which can reduce later switching time. The first use of a model can show an "Optimizing…" message while Saywave prepares it for your Mac.

Speaking a language other than English? Pick it from the Language submenu — this switches Saywave to a multilingual engine (Parakeet or Apple Speech) that currently covers English, German, French, Spanish, and Italian.

3. Cleanup modes — Raw, Clean, AI Rewrite, Translate

The Mode submenu controls what happens to your words before they're pasted:

• Raw — the speech model's transcript without cleanup.
• Clean — light polish: filler words removed, punctuation and capitalization fixed. This is the default.
• AI Rewrite — a full rewrite in a chosen style (e.g. email, message, prompt), using an on-device or your own AI model.
• Translate — speak in one language, get text in another.

Turn on Adapt mode to app in Settings → Cleanup & AI and Saywave will pick the mode for you based on what's frontmost — Email in Mail, Message in Slack or Messages, Prompt in your terminal or editor — so you never have to switch modes by hand.

Spoken formatting commands

Say "new line", "new paragraph" or "bullet point" while dictating and Saywave inserts the break or bullet instead of the words — in English and in your dictation language (German, French, Spanish and Italian are built in). The command parser is designed to distinguish commands from normal speech, but you should review the result. Commands work in every mode except Raw, and can be turned off in Settings → Cleanup & AI.

4. Rewrite Selection

Already have text written and just want it cleaned up or rewritten? Highlight it anywhere, then choose Rewrite Selection from the menu bar (or its shortcut). Saywave reads the highlighted text, runs it through your current mode, and pastes the result back in place — no need to dictate it again.

5. Dictionary

Settings → Dictionary teaches Saywave how to spell names, technical terms, or anything it tends to mishear. Add a word on its own to teach the correct spelling, or fill in both fields to replace one spoken phrase with another every time.

On Whisper models your terms can guide recognition itself. Apple Speech and Parakeet don't offer that hook yet — there your replacement rules are applied right after transcription instead.

6. History & Statistics

Every dictation is saved locally under Settings → History — click any entry to copy it again. Statistics tracks your total words dictated, speaking time, and words-per-minute over time, entirely on your device.

7. Meeting Capture

Saywave can record both sides of a call — your microphone and the other participants through your Mac's system audio — and turn it into a searchable, speaker-labeled transcript. Summaries and action items can use an on-device backend or, if you selected one, OpenAI or Anthropic.

Starting a capture

Open the menu bar's Meeting submenu and choose Start Meeting Capture. The Stop item shows the running time so you always know a capture is live. Choose Stop Meeting Capture when the call ends.

Auto-capture (optional)

Turn on Auto-Capture in Zoom, Teams & FaceTime in the same submenu and Saywave starts recording automatically the moment one of those apps opens, then stops the moment it quits — no clicking required. It is off by default. Enable it only if your meeting workflow ensures every participant receives advance notice and any required consent is obtained before automatic capture starts.

Permissions

The first time you start a capture, macOS asks for Screen Recording access (this is the system permission that also covers system-audio capture — no video is ever recorded or stored). You can grant it during onboarding or the first time you start a meeting; macOS requires a relaunch of Saywave after granting it before capture will work.

Reading & exporting a transcript

A few moments after a capture ends, its transcript appears automatically in Settings → Meetings — no action needed. Each meeting shows the source app, date, duration, and number of voices detected. Click a meeting to read it: your side of the conversation is labeled You, and each other participant is labeled Speaker 1, Speaker 2, and so on, in the order they first speak.

Use the search bar to find a meeting by app, date, speaker, summary, or anything that was said. From an expanded transcript, Copy Transcript copies the whole conversation with timestamps; Copy One Voice copies just one speaker's lines — handy for pulling only your own notes or only what someone else said. Prefer a file? Export saves the whole note — summary, action items, and transcript — as a Markdown or plain-text file.

Summaries & action items

A short summary and a list of action items can appear at the top of each meeting. With an on-device backend, the transcript remains on the Mac for summarisation. If you have selected OpenAI or Anthropic, generating a summary sends the meeting transcript and instruction to that provider using your API key. Summaries generate automatically when you're using the on-device engine; if you've set up a cloud key instead, or configured AI after a meeting was recorded, a Generate Summary button creates one on demand. No AI configured yet? The meeting shows a gentle hint pointing you to Settings → Cleanup & AI — never an error. Copy Summary copies just the summary and action items.

Naming speakers

Each meeting starts by labeling voices You and Speaker 1, Speaker 2, and so on. Click any speaker chip above the transcript to give them a real name — it's saved with the meeting and flows through the summary, action items, and every export.

Turn on Remember voices across meetings at the top of the Meetings pane and Saywave learns each voice as you name it, then labels the same person automatically the next time they join a call. These voice fingerprints live only on your Mac and never leave it; manage or forget them any time from the same panel. It's off by default.

8. Troubleshooting

Dictation isn't typing anything

Check the menu bar status line first. "No audio — check microphone" means your microphone captured only silence (often a muted headset mic). Saywave always records the microphone selected in macOS Sound settings; the Input submenu shows that same selection and changes it system-wide, so pick a working device there and dictate again.

"Grant Accessibility" keeps appearing

Saywave needs Accessibility permission to paste text at your cursor. Grant it from the menu bar's Grant Accessibility… item or in System Settings → Privacy & Security → Accessibility, then try again — no relaunch needed.

A model takes a while to load the first time

The first use of a model can include an optimization step for your Mac's Neural Engine. Duration depends on the model, hardware, memory pressure, and network if a download is needed.