# Features — every WrapScribe tab and feature explained

> Markdown twin of https://wrapscribe.com/docs/features — the canonical page.

Documents WrapScribe 1.9.1.

WrapScribe lives in your menu bar. You hold a key, talk, and the words arrive at your cursor. Everything below is what the app builds around that: the Dashboard, the meeting recorder, the rewrites you can ask for out loud, and the parts that appear only when something goes wrong.

Words in quotes are the app's own. Anything marked (Pro) needs a Pro or Max licence.

## How a dictation works

### Starting and stopping

Hold the push-to-talk key and talk; let go and the text arrives. On a new install that key is `Fn`.

For anything longer there is the hands-free shortcut — `Cmd ⌘` + `Fn` by default — which starts recording on one press and stops on the next. `Esc` cancels without pasting. If your mouse has a thumb button, "Enable mouse dictation" turns it into a push-to-talk key too. Recording never starts by itself.

*Mock-up: Four key-caps in a row — Fn held to talk, Cmd ⌘ + Fn to toggle, Esc to cancel, and a mouse with its side button lit.*

[Set your shortcuts](/docs/settings#shortcuts-dictation)

### What the overlay shows

A dictation moves through four states and the overlay tracks them: getting ready, recording, transcribing, and a last moment of feedback if there is anything to say.

The default is the "Floating pill", a dark capsule you can park anywhere on screen. It draws a live waveform while you talk, then a spinner; in hands-free mode, hovering it reveals two controls — ✕ discards the recording, ✓ finishes it early. The alternative is the "Minimalist menu-bar band", a slim strip beside the notch that shows the waveform and nothing else and "Never covers your app's toolbar". A third option is no overlay at all: "Record with the menu bar only" leaves a red dot and a running timer in the menu bar to do the whole job.

*Mock-up: The Floating pill in four states side by side — initializing, recording with a live waveform, transcribing with a spinner, and a feedback line.*

[Choose your overlay](/docs/settings#overlay-recording-overlay)

### The chips that stay

Three messages do not fade on their own; they wait for an answer, in whichever overlay you picked. An update chip reads "WrapScribe 1.8.8 is ready" and offers "Update" or "Later". A permission chip appears when macOS has taken something away — the microphone, Accessibility, Input Monitoring — and offers "Open Settings" or "Later". When your licence rises to a bigger plan, a chip says "Now on Pro — Meetings and more unlocked" with a single "OK", because there is nothing to defer about good news.

*Mock-up: The pill wearing its update chip — headline on the left, an Update capsule and a Later link on the right.*

### Where the text goes

When you stop talking, WrapScribe looks at what was in front of you and picks one of three endings. An editable field was focused and you are still in the same app: the text is typed at your cursor. WrapScribe's own window is frontmost: the transcript goes quietly onto the clipboard. Neither: you get the copy card — or, if Accessibility itself has gone, the card below.

The paste borrows your clipboard for a moment and puts back what was there before, unless you switch that off.

### The copy card

The copy card leads with "Select a textbox first, then dictate", shows the transcript in full, and has one primary button — "Copy", which Return also presses.

Underneath, quietly, sits "Always paste in ‹App›". Some apps — GPU-rendered terminals, and Electron or browser shells shipped with accessibility switched off — never tell WrapScribe a field is focused, so the card appears in them every time. Accepting adds that app to a list, and WrapScribe pastes there without asking from then on.

*Mock-up: The copy card — WrapScribe mark, the headline, a scrollable transcript, a white Copy button, and the quiet “Always paste in Dia” line beneath.*

[The list of taught apps](/docs/settings#shortcuts-always-paste)

### When Accessibility goes missing

macOS ties Accessibility to the exact app it granted it to, so an update replacing WrapScribe on disk can quietly revoke it, and nothing can be typed at your cursor until it is back. The card that appears says so — "Accessibility permission was lost after updating" — and adds a "Re-grant" button beside "Copy". Your transcript is still there and still copyable; only the paste is blocked.

[The banner in Settings](/docs/settings#shortcuts-banner-accessibility)

### What cleanup does, in order

Between the microphone and your cursor the text runs through a fixed sequence. Knowing the order explains most surprises.

1. **Spoken self-corrections** resolve first, so a spoken "no, no" that walks back the last word is honoured before anything else touches the text.
2. **Spoken emoji** become emoji, if you asked for that.
3. **Filler words** are stripped deterministically, with or without Apple Intelligence.
4. **Your corrections** are applied: the "Exact Corrections" you typed, plus pairs learned from fixes you made by hand.
5. **A structure you asked for out loud** is detected when it opens the dictation followed by a colon and the items ("give me three points: ship the build, test the installer, tell the team") or closes it right after them ("…put this in three bullet points"), and the request itself is removed rather than becoming the first bullet. The same words anywhere else, or a prompt such as "give me five points about the launch", are typed as you said them.
6. **Spoken numbers** become digits, so "Grok four point five" is a version.
7. **Cleanup** runs. "Basic" is the instant deterministic pass; "Smart" sends the text to Apple's on-device model with the active app, window title, any selected text, your dictionary terms, output language, instructions and that app's writing style. If the model is unavailable or slow, Basic takes over and the app says so rather than hiding it. So does an answer that leaves out most of what you said, after one more try.
8. **Snippets expand** — after cleanup, so tidying can never break a trigger.
9. **New words are noticed** for the dictionary.

"Exact" mode returns before steps 1 to 7 and step 9: it keeps the words you spoke, with only the minimum formatting needed to insert them. Snippets still expand.

*Mock-up: A before-and-after pair — a raw transcript full of “um” and a repeated phrase on the left, the cleaned sentence on the right.*

[Cleanup settings](/docs/settings#cleanup-mode)

### History and recovery

The last 20 dictations stay on this Mac, transcript and all, so you can reread or re-copy them. Nothing is uploaded.

A dictation you cancelled keeps its audio and shows up as an italic line — "This transcription (0:14) was dismissed." — with a "Recover" button that transcribes it after the fact. The pill puts a shorter version in its last feedback moment — "Dismissed — click to recover" — and clicking it opens the Dashboard, which starts on that history.

[History and auto-delete](/docs/settings#general-history)

## The Dashboard

The Dashboard is the window: 960 × 620, resizable, with a sidebar down the left. Open it from the menu bar with "Open WrapScribe…", or by launching WrapScribe from Applications or Spotlight. It has no keyboard shortcut of its own.

Six tabs sit in the sidebar — Dictation, Dictionary, Snippets, Meetings, Transforms, Styles — with one pinned row at the bottom: "Settings". Three tabs are Pro; on Basic they stay visible and explain themselves rather than vanishing.

*Mock-up: The Dashboard window — sidebar with the WrapScribe mark, six tabs with Dictation selected, and the Settings row pinned at the bottom.*

### Dictation

The first tab greets you by first name — "Welcome back, ‹name›" — and gives you two columns.

On the left, under a "RECENT" label, is your history with a "Search" field above it; each row is a time and a transcript, and clicking anywhere on it copies the text — except a cancelled one, which offers "Recover" instead. A range picker in the header narrows everything to "All", "Today", "3d", "7d", "30d" or a "Custom" pair of dates.

On the right are four numbers — total words, dictations, a day streak with a flame beside it, and your longest streak. Words and dictations follow the range; streaks are always lifetime. On a Mac that has just signed in the four tiles spin and the card reads "Loading your stats…" until the account's totals arrive. The line beneath says where the other Macs' figures come from: "Other Macs add their daily counts from the day they were activated."

Below that, a "Microphone" card names the device actually listening and lets you change it, and an "On-device" card states the deal: "Transcription runs on this Mac and your words never leave it."

*Mock-up: The Dictation tab — recent transcripts on the left, four stat tiles, the Microphone card and the On-device card stacked on the right.*

### Dictionary

The Dictionary teaches WrapScribe "the names, products, and acronyms it should always spell right." These are hints for the speech model, not find-and-replace: a word here makes the right spelling likelier, it does not force a substitution. For substitutions, use "Exact Corrections".

Type into "Add a word or phrase…" to add one. Rows can be starred, switched off, or deleted. Words WrapScribe noticed by itself arrive wearing a "suggested" chip and an "Add" button, and become active once they have appeared in three successful dictations.

The dictionary never syncs, because it is fed from what you actually said. It belongs to the account that built it, so "Export…" and "Import…" — which move it, and your snippets, to another Mac — are greyed until you sign in.

*Mock-up: The Dictionary tab — an add field, a search field, and rows of saved words, one starred and one wearing a “suggested” chip.*

[Automatic learning and corrections](/docs/settings#cleanup-automatic-learning)

### Snippets

"The stuff you shouldn't have to re-type. Say a trigger and WrapScribe drops the text in."

A snippet is a "Trigger" and what it "Expands to". Matching is whole-phrase and ignores case and accents; where two triggers could match at the same spot, the longer one wins. Expansion happens after cleanup. A new install ships with two examples, "my email" and "intro email", to show the shape.

*Mock-up: The snippet editor sheet — a Trigger field reading “my email” and an “Expands to” box below it.*

### Meetings (Pro)

The Meetings tab lists everything recorded or imported. Each row carries a status chip — "Recording", "Processing", "Ready" or "Failed" — plus an "Imported" chip for anything that came from a file. Opening a row replaces the list with the meeting: the title (click to rename), an audio player, and three tabs.

"AI Notes" is an editor seeded with what the on-device model wrote; click anywhere in it to edit, and your version is what exports from then on. "Regenerate" runs the model over the transcript again. "Full Transcript" is the timed record — `[m:ss]`, the speaker, the line — with a search box that counts and steps through matches; the two speaker labels start as "You" and "Others" and are editable. "Chat" answers questions about the meeting from its own notes and transcript.

*Mock-up: The meeting detail view — back chevron, title, status chip, the audio scrubber, the three tabs with AI Notes selected, and editable notes below.*

### Transforms (Pro)

A transform is a named rewrite you run on your last dictation by voice, or on any selected text with a keyboard shortcut. Two ship with the app: "Polish" tightens grammar and flow, removes fillers and repetition, and keeps your meaning, tone and hedges exactly; "Prompt" rewrites what you said as a prompt for an AI assistant — a "Goal:" line and one bullet per remaining condition.

"Create New" opens an editor with a "Name (what you say after the wake word)", an "Instruction", and a "Keyboard shortcut (optional)". That last one is the interesting part: "Select text in any app, press this shortcut, and WrapScribe runs this transform on the selection." A shortcut clashing with a dictation shortcut or another transform is refused with a reason.

By voice a transform answers to its own name, to "‹name› that", to "‹name› this", and to "apply ‹name›" — as the tab's own subtitle puts it, "Hey WrapScribe, polish that".

*Mock-up: The Transforms grid — Polish and Prompt cards each marked “Built-in”, a user transform with a shortcut chip, and the Create New button.*

### Styles (Pro)

The Styles tab is titled "Writing Styles", and its job is one line: "Teach WrapScribe how to write in each app."

"Add app" offers a short curated list — Slack, Mail, Apple Notes, Messages, Discord, Microsoft Teams — plus "Other app…" for anything in your Applications folder. Adding one asks the same question, "How should WrapScribe write here?"; the card it leaves behind is your answer in an editable box, saved as you type.

That instruction reaches Smart Cleanup whenever you dictate into that app, and the selection assistant when you rewrite text there. If a licence later says Basic, saved styles stay put and simply stop being applied.

*Mock-up: Two Writing Styles cards, Slack and Mail, each with its app icon and an editable instruction box.*

[The formality dials](/docs/settings#cleanup-writing-style)

### What Basic sees

On Basic the three Pro tabs keep their place, and their own title and subtitle stay at the top; the page below becomes a locked card: the feature's name, what it does, "Part of Pro — one purchase, up to 3 Macs." and an "Upgrade to Pro" button. Underneath is the sentence that matters most — "Dictation, cleanup, your dictionary, snippets, overlays and history are on your plan already and keep working exactly as they do now."

*Mock-up: The locked Meetings page — the tab's own icon with a small lock, the feature name, the Pro line and an Upgrade to Pro button.*

## Talking to WrapScribe

### Ask WrapScribe (Alpha)

Inside an ordinary dictation, start with "Hey WrapScribe" and the rest becomes a request rather than text to type: ask a question, generate something, or change what you just dictated — "make that formal."

WrapScribe uses the active app and recent text as on-device context, removes the trigger phrase, and pastes the answer. A shorter trigger — plain "WrapScribe" — is available and off by default. With "Use on-screen text as context" on, a request like "reply to this email" reads the visible window entirely on-device, through accessibility or a window snapshot when Screen Recording is already granted. The app is candid about the state of it: "Inline AI is in alpha."

*Mock-up: An annotated transcript — the spoken sentence with “Hey WrapScribe” underlined as the trigger and the rest marked as the request.*

[Ask WrapScribe settings](/docs/settings#shortcuts-ask)

### Edit Mode

Edit Mode is the other way to rewrite: select text, hold your dictation shortcut, and say what should change. "Transform highlighted text with a spoken instruction instead of dictating over it."

It is off until you turn it on. "Automatic" lets your normal dictation shortcut do the job whenever text is selected; "Manual" leaves dictation alone and asks for one extra modifier held alongside — `Cmd ⌘`, `Ctrl ⌃`, `Option ⌥` or `Shift ⇧`.

[Edit Mode settings](/docs/settings#shortcuts-edit-mode)

### Spoken commands

A few phrases do something instead of being typed. End a dictation with "press enter" and WrapScribe removes those words, pastes the rest, then presses Return. Say "scratch that" or "delete that" as the whole dictation and it deletes what it just pasted, as long as that text still sits right before your cursor. Spoken emoji names become emoji when you ask for them: say "thumbs up" to get 👍, while everyday words like "fire" or "heart" only turn into emoji when you say "fire emoji".

[Clipboard and command settings](/docs/settings#shortcuts-clipboard)

### Voice Macros

A voice macro is a command and a block of text: say the command, get the text. It "Bypass post-processing and immediately paste your predefined text" — no cleanup, no rewriting, exactly the characters you saved. Macros are made in Settings, with a "Voice Command (What you say)" and a "Text (What gets pasted)".

[Voice Macros](/docs/settings#shortcuts-macros)

### Transforms by voice

Any transform's name is a spoken command. After the wake phrase, "polish", "polish that", "polish this" and "apply polish" all run the same thing. Names match whole, so "polish that thing up" is treated as ordinary speech.

## Selection assistant (Pro)

Select at least three characters in an editable field and a small blue "+" bubble appears beside it. It is off until you turn it on.

Clicking it opens a panel with two tabs. "Instant" lists the one-click rewrites: "Grammar" ("Fix spelling, grammar and punctuation"), "Polish" ("Improve clarity, flow and tone"), "Translate" with a "Change" button for the target language, "Custom request…" for a one-off instruction, and under "Other modes", collapsed until you open it, every transform — the two built in and any you have made. "Chat" is a conversation about the selection — "Ask for a change…" — where the newest answer carries an "Apply" button bound to `⌘↩`.

The result is pasted over your selection. When that is not possible the panel says "Copied — paste it yourself" and leaves the text on the clipboard. When an app returns no selection, WrapScribe wakes its accessibility tree once and looks again; if that comes back empty too, no bubble appears.

*Mock-up: The blue + bubble beside a selected sentence, with the assistant panel open on its Instant tab.*

[Turn the assistant on](/docs/settings#assistant-selection)

### Translate

Translation runs on-device with Apple's translation models. The first use of a new language may ask to download it — macOS shows a one-time prompt. Pick the target from the assistant's Translate row; a pair Apple does not support falls back to the on-device language model.

## Meetings (Pro)

### How a meeting is noticed

WrapScribe watches two local signals and nothing else: which app is using the microphone, and which meeting is open — in the app in front of you, or in a browser it re-checks every fifteen seconds.

A call in Zoom, Microsoft Teams, Webex, FaceTime, Slack or Discord counts. So does one open in any browser — Google Meet, Zoom, Microsoft Teams, Webex, Whereby, Jitsi Meet, Amazon Chime, Around, Gather or Discord — recognised from the tab's address, and for most of them from its title too. The browser list is discovered from whatever your Mac opens links with, so Safari, Chrome, Arc, Dia, Brave, Edge and Firefox are all covered.

By default the prompt arrives as soon as you open a meeting. Turn "Offer to record as soon as I open a meeting" off and it waits for the microphone to go live — arriving a beat after you join, and never at all if you stay muted. The Slack and Discord apps never trigger the early prompt: they sit in the foreground all day, so "frontmost" says nothing about a call. A Discord call in a browser still does.

Recording never starts on its own. Offering is the only thing detection can do.

### Recording one

The prompt is a small dark card headed "Record this meeting", with the source and app underneath, that app's icon on the left, and a blue "Start recording" button Return also presses. A chevron beside it opens the audio choice — "Meeting app audio only" or "All system audio" — and one more item, "Don't ask for ‹App› again".

Once recording, the card becomes a live pill: the WrapScribe mark, a pulsing red dot, a running timer, a minus disc that minimizes it, a red "Finish" button, and a ✕ on hover that discards the recording.

Two tracks are captured — your microphone as "You", the system audio the others arrive on as "Others" — through a Core Audio process tap. macOS asks for System Audio Recording permission once, at the first recording; without it the recording does not start at all, and WrapScribe says so.

*Mock-up: The “Record this meeting” prompt card above the live recording pill with its timer, Minimize, Finish and discard controls.*

[Meeting settings](/docs/settings#meetings-notes)

### AI Notes

When a recording finishes, Apple's on-device model writes the notes: a title, a summary of two to four sentences, key points, decisions the participants agreed on, and action items each starting with a verb. Long calls are summarised in chunks. "Regenerate" re-runs it over the same transcript. The "Chat" tab needs Apple Intelligence and says so when it is unavailable rather than pretending.

### Import, copy, export, delete

"Import audio or video…" takes a recording you already have — WAV, M4A or MP3, or the audio track out of an MP4 or MOV — and runs it through the same transcription and notes. Imported meetings are marked, because they have only one speaker.

"Copy" copies whichever tab you are looking at. "Export" offers the AI notes or the full transcript as "Plain text (.txt)", "Markdown (.md)" or "PDF (.pdf)", and the same menu sits on every row in the list. "Delete meeting" removes its transcript, notes and audio from this Mac.

## Speech engines and languages

Transcription happens on this Mac, always. Five engines are offered, and the picker leads with what each one is rather than what it is called.

- **Built into macOS** — Apple Speech. "No download, and the fastest to start." Speed 5/5, accuracy 3/5. One language at a time.
- **Smallest download** — Whisper Base, 145 MB. "Quick and light, but rough on accents, names and anything noisy." Speed 5/5, accuracy 2/5.
- **Balanced** — Whisper Small, 483 MB. "Good enough for everyday dictation, in almost any language." Speed 4/5, accuracy 3/5.
- **Most accurate** — Whisper Large v3 Turbo, 1.6 GB. "Worth the wait for strong accents, hard audio and Indic languages." Speed 2/5, accuracy 5/5. No live preview — it cannot re-decode fast enough to keep up with a speaker.
- **Fastest** — Parakeet TDT v3, 483 MB. "Keeps up while you talk, with a live preview. 25 European languages." Speed 5/5, accuracy 4/5.

Above the list you name the languages you speak. Leave it empty and it reads "Detect automatically". Name one and it is used for every dictation, the most accurate option when you only speak one. Name several and the model detects per dictation — except Apple Speech, which "can only handle one language at a time".

Whisper covers 99 languages; Parakeet covers 25 European languages, and no Indic languages at all. If your selection and your engine disagree, a banner names the problem and offers to switch or download the one that fits. Downloaded models "are stored in Application Support and never leave this Mac."

Output Language is a separate question: Smart Cleanup translates the finished text into the language you pick; Exact mode always keeps the spoken one.

**Which one to pick.** Parakeet — the row marked **"Fastest"** — is the recommended model for every use case. It is the quickest at keeping up with speech, has the fastest-refreshing live preview of any model, and at 4/5 accuracy it is beaten only by the 1.6 GB Whisper Large v3 Turbo, which is far slower. Download it once (483 MB) from Settings → Language & Model and stay on it; only a language it does not cover sends you to a Whisper model.

*Mock-up: The Speech Model list — five radio rows with persona titles, star ratings, download sizes, and one row mid-download.*

[Languages and models](/docs/settings#model-speech-model)

## Around the app

### The menu bar

The menu-bar icon is the front door. Its menu opens with a disabled status line — "WrapScribe — Ready", or "Dictating  0:12", "Recording meeting  4:03", "Transcribing…" — then, in order: an update row when there is one ("Update to WrapScribe 1.8.8…"), "Start Dictating" or "Stop Dictation", "Record a Meeting" or "Stop Recording" on Pro, "Open WrapScribe…", "Settings…", "Re-run Setup…", and "Quit WrapScribe".

The icon has four looks: the WrapScribe mark at rest, a red ● with a running timer while anything is recording, a "…" while a transcript is still finishing, and a "↑" while an update waits. Hovering the last explains it: "WrapScribe 1.8.8 is ready to install".

*Mock-up: The menu-bar dropdown — status header, update row, Start Dictating, Record a Meeting, Open WrapScribe…, Settings…, Re-run Setup…, Quit.*

[Hide or recover the icon](/docs/settings#general-app)

### Updates

WrapScribe checks for a new version once an hour, and on demand. A fresh release is held back from an automatic offer for its first six hours; "Check for Updates Now" ignores that hold and offers it straight away. When one exists you meet it in up to three places: the menu-bar row, a chip on the overlay after a dictation, and Settings → General.

Asking for it opens a dialog headed "A New Version is Available" with four answers — "Download Update", "What's New", "Remind Me Later" and "Skip This Version". The install happens in place and the app relaunches itself. A backup is written before it does.

[Update settings](/docs/settings#general-updates)

### Backups and export

WrapScribe backs up your dictionary, snippets, stats and learned corrections once a day at launch and again before an update, keeping the newest seven in Application Support → WrapScribe → accounts → your account → Backups. If that account ever comes up with the dictionary, snippets and stats all empty, the newest backup is put back automatically and Settings → Account says so afterwards. An account is only ever handed its own backups.

That is a safety net, not a migration tool. To move to another Mac use "Export…" and "Import…" on the Dictionary or Snippets tab: one file carries both, plus stats and corrections, and importing never overwrites: words, snippets and corrections merge, and stats are adopted only onto a Mac that has none of its own. Both buttons need an account: signed out they are greyed, with "Sign in first — your dictionary and snippets belong to your account." beside them.

### What the account carries

Your account carries two things: your word and dictation counts from each Mac, and the preferences worth having on a new machine. Nothing you said travels — not transcripts, not audio, not your dictionary. Every Mac reports its lifetime words, dictations and longest streak with each sync, so signing in on a new machine fills the Dashboard from the account's real totals rather than from the day you installed it. Your history, meetings, dictionary and snippets are filed under your account on each Mac — a different licence key starts clean, and signing back in brings yours back — but they stay on the Mac that made them.

[What syncs across your Macs](/docs/settings#sync)

### What leaves this Mac

Transcription, cleanup, meeting notes, translation and every rewrite run here. Three things go out over the network: the licence and sync server, the update manifest, and the speech model you download the first time you pick one. That is the whole list. WrapScribe also keeps its own overlays out of screen recordings by default, so a shared screen never shows the pill, the copy card, the meeting pill or the selection panel.

### Plans

WrapScribe is bought once, on three plans. **Basic** is "1 Mac" — dictation and everything around it: cleanup, the dictionary, snippets, every overlay, history and recovery. **Pro** is "up to 3 Macs" and adds four things: "Meeting recording", "Selection assistant", "Transforms" and "Per-app writing styles". **Max** is "every Mac you own", with Pro's features.

A plan change is picked up on the app's next licence check, without a relaunch, and announced with a chip on the overlay. There is no seat screen: you free a Mac by signing out on it, in Settings → Account.

### Every keyboard shortcut

| Key | What it does |
|---|---|
| `Fn` (hold) | Push to talk — record while held |
| `Cmd ⌘` + `Fn` | Hands-free mode — press to start, press to stop |
| `Esc` | Cancel the dictation without pasting |
| Unset | Paste again — re-insert your last dictation at the cursor |
| Your choice | Run one transform on the selected text, per transform |
| `⌘,` | Open Settings |
| `⌘Q` | Quit WrapScribe |
| `⌘↩` | Apply a rewrite in the assistant's Chat tab |
| `⏎` | Press the default button — Copy on the copy card, Start recording on the meeting prompt, Save in an editor |

Every dictation shortcut can be re-recorded or switched off, except Cancel, which always keeps a key.

[Change any of these](/docs/settings#shortcuts-dictation)