What is new
Every build that has actually shipped, newest first. The list is generated from the same update feed the app reads, so a release cannot appear here without having been published — or be published without appearing.
Updates arrive through the updater inside the app: Voicecape checks once a day and sends nothing but a version number. You can also download the current build directly.
Nothing shipped in that area yet.
1.5.7 — Release temporary speech models after recording
When Qwen is selected for dictation, the temporary local transcription model is released after a recording finishes, fails or is cancelled. Longer recordings reuse it between processing windows.
- Your selected dictation model stays ready. Audio processing remains on your Mac.
1.5.6 — Smoother typing and a lighter workspace
Keyboard shortcut monitoring now runs independently of the app’s interface, so interface work does not hold up ordinary typing in other apps.
- Fixed an input suppression setting after paste and improved shortcut release and cancellation handling.
- Moved local database work into the background and reduced repeated search requests.
- Local speech models now prepare in the background, with cancellation and settings changes preserved during loading.
- Large recording lists load visible rows as needed. Microphone activity no longer rebuilds saved recordings.
- Restoring speaker status no longer loads the full transcript. Audio stays on your Mac.
1.5.5 — A clearer desktop workspace
The Voicecape identity now appears consistently in the app icon, sidebar and setup. Recordings use a compact list with transcription, speakers and summary actions in the selected recording’s details.
- Browse Knowledge, Dictionary, Snippets and History with quieter toolbars and editors that open when needed.
- Find settings by category, with current values and contextual help.
- Audio stays on your Mac. AI summaries still require your consent to send text.
1.5.4 — Separate speakers on your Mac
Identify anonymous speakers in locally transcribed recordings up to 45 minutes long. Voice analysis runs on this Mac; audio and voice embeddings are not sent to a server.
- Review Speaker 1, Speaker 2 and uncertain or overlapping speech in Recordings. Speaker labels are estimates, not verified identities.
- Choose to refresh an AI summary with speaker-labelled text after granting consent. Your existing summary stays unchanged until you request an update.
- Cancel analysis at any time. Closing the speaker window lets it continue in the background.
1.5.3 — Meeting notes open again, recording language fixed
Meeting notes recorded with earlier versions open again. A note without a private local transcript simply shows no transcript section.
Local transcription and AI summary requests now send a plain language code. Recordings made under a region-specific language setting such as Korean (Korea) no longer fail.
Recordings that belong to another account on this Mac stay hidden and cannot be played, revealed, resumed or deleted until that account signs in.
Cancelling after a transcript has already finished no longer blocks its summary.
1.5.2 — Cancel local model installation reliably
Cancelling a Qwen model download during the final file verification now stops installation. A model already installed on this Mac stays intact.
1.5.1 — Open the local library after a trial ends
When the trial has ended or setup is still shown, you can open the Local library without activating an account to read and copy items already visible on this Mac. This read-only view does not start new recordings, AI processing, or sharing; those actions keep their existing access checks.
1.5.0 — Knowledge and local recording transcripts
Knowledge brings desktop notes and recordings into one place to browse and organize. After a Mac recording finishes, you can transcribe its saved audio on this device. That local transcription does not send the recording audio to a speech provider. It can resume from the saved recording if processing is interrupted.
AI summaries are optional and require your consent. When you request one, the completed transcript text is sent for summarization; the recording audio stays on your Mac. Workspace administrators can set the summary depth. Review the transcript and summary before relying on them: local transcription does not identify speakers automatically.
After reviewing a summary, you can explicitly share it to your workspace Knowledge. People and connected MCP tools with permission can read that shared summary. If you also select Slack, it receives a short preview and a link; selected Notion receives the full reviewed summary. Creating a draft does not share it.
- Dictation feedback is quieter. Stopping capture alone makes no generic sound; a short cue follows a successful insertion, while clipboard, refusal, and failure outcomes have distinct restrained cues.
- On Apple Silicon Macs running macOS 14 or later, you can optionally download and select the Qwen local recognition model in Settings. It is not downloaded automatically. The existing Whisper model remains available, including on supported Intel Macs, and selecting Qwen does not silently switch to cloud recognition.
1.4.1 — Recording sound fixes
Recordings now keep the other side of the call at its real volume. Turning on the microphone used to lower the sound of other apps on the Mac while recording, so the meeting or video was quieter both for you and in the recording. That no longer happens.
- Playback in Recordings now plays the microphone and the Mac’s sound together in both ears. Before, your voice came from the left and the other side from the right.
- The first second of the Mac’s sound is no longer missing at the start of a recording.
- The “Microphone audio stopped” notice no longer appears at the start of a normal recording while the microphone is still starting up.
1.4.0 — Recordings and snippets
Voicecape can now record meetings and calls on your Mac. Open Recordings from the main window or from Settings, confirm that everyone being recorded has been told and has agreed, and press Start recording. The microphone and the system audio of the Mac are recorded together, and the menu bar shows how long the recording has been running. You can pause and resume, and a recording interrupted by quitting the app is recovered the next time the app opens. Finished recordings are saved on this Mac. From Recordings you can play one, show it in Finder or delete it, and each one also appears on the board as a note. Recording requires macOS 14.4 or later, and dictation is paused while a recording is running.
If you are signed in and your account has transcription minutes available, you can also turn on live transcription when you start a recording. While you record, a mono copy of the audio is streamed for transcription to the provider named in the privacy policy. The audio is not stored in the cloud, and the original stays on this Mac. The transcript is added to the note for that recording, and when it is complete a summary is requested automatically. An interrupted transcription can be resumed from the audio kept on this Mac.
Snippets is a new pane next to Dictionary in Settings. Save a signature, an address or a line you type every day under a phrase, then say that phrase on its own and the saved text is typed exactly as written. Only the phrase spoken by itself triggers a snippet, so ordinary sentences are never replaced, and the snippet text stays on this Mac even with Polish with AI turned on. Each snippet shows how many times it has been used.
- Say “scratch that” or “scrap that” on its own to remove the text the last dictation typed. It works for two minutes after that dictation and only while the same app is still in front. It is ignored in Verbatim mode and in code editors and terminals, and if there is nothing to remove, nothing is typed.
- With Polish with AI on, the new Tone of the finished text setting chooses how dictation into chat and writing apps reads: left as you said it, casual, polite or formal. Writing apps default to polite and chat apps keep your wording. It applies only to apps filed as chat or writing, so code editors, terminals and other apps are left as dictated.
- The board has a New note field, so a note can be typed as well as dictated.
- Backups now include your snippets. A backup made with this version can be imported by 1.4.0 or later. Earlier versions decline the file instead of importing only part of it.
- Fixed a crash that could occur when the dictionary changed while a dictation was being processed.
- Onboarding, the trial screens and the microphone permission prompt no longer describe Voicecape as a one-time purchase that runs only on this Mac. They now say that it is a subscription, that Cloud dictation sends audio for recognition when it is turned on, and that the speech model inside the app takes over when you are offline.
1.3.3 — Quick note
Press the capture shortcut and say something. It lands on a board inside Voicecape, and you do not have to open the app first or find a window to type into. Both shortcuts — one to capture, one to show the board — are in Settings → Quick note, and either can be turned off.
Notes stay on the Mac that took them. Nothing on the board is deleted on its own: you can open a note to edit it, copy it out, or delete it yourself. Dimming after a week, two weeks or a month only marks a note as having sat a while — it does not remove anything.
1.3.2 — Take your dictionary with you
Settings → Backup writes your settings and your personal dictionary to a file you choose, and reads one back on another Mac. Importing only adds — nothing already on this Mac is removed, and importing the same file twice does not duplicate anything.
Your transcripts are not in that file. They stay on the Mac that made them, which is the point of this product; the file carries the vocabulary you built up, which is the part that is painful to lose. Machine-specific things are left out too — a microphone that only exists on the old Mac should not follow you to the new one.
1.3.1 — Free a Mac yourself
If your licence is on all three Macs and you are retiring one, you no longer have to write to us and wait. Open Settings → Licence on the Mac you are letting go and press Release this Mac — the slot is free straight away, and the same licence key brings it back whenever you want. Email is still the answer for a Mac you no longer have, because a machine that is gone cannot release itself.
Settings can also give a tone to apps that are not in the built-in list. After you dictate somewhere, that app appears at the top of Tone per app and you can file it as code, terminal, chat or writing. Until now only sixteen apps could be classified and everything else was treated as unknown, which quietly meant the most cautious clean-up with no way to say otherwise.
1.3.0 — Accounts
You can now create a Voicecape account with your email address, or sign in with Google. An account is not a licence and it is not required to dictate — it is a way back in that does not depend on finding a licence key in an old inbox. (An earlier version of this note said an account carries your settings, dictionary and history between Macs. It does not, and we should not have written it: nothing is synced to a server. 1.3.2 adds a backup file that moves your settings and dictionary yourself — your transcripts stay on the Mac that made them.)
Signing in has its own shape now: you pick sign in or create account once at the top, and the screen below follows that choice instead of offering three buttons that all look equally likely. If you forget your password we email you a link, which lands on a page on this site where you set the new one.
1.2.9 — Fewer ways to lose a device slot
This Mac now settles on its identifier the first time it needs one, instead of only when an activation succeeds. A failed attempt on a bad connection used to leave nothing behind, so the next press looked like a different Mac to the server and quietly used up another of your three slots.
Two smaller fixes in the same area: cloud dictation now refreshes its credentials when you come back to the app even if you never turned on clean-up, and a trial that runs out while the app is open is noticed there and then rather than on the next launch.
1.2.8 — Activated licences now actually unlock the app
A licence could be activated successfully — stored, signed, ninety days of offline cover — and the app would still show the trial-ended screen, because the check that reads that licence could not see the signature it needed. It can now, and an activated Mac unlocks immediately.
1.2.7 — Licence messages in your language
When activation is refused, the reason now comes from the app in the language you use, instead of an English sentence relayed from the server. And if a licence is already on all its Macs, the message tells you what actually works: write to us and we release a slot the same day.
Repeat presses of Activate no longer make a second activation of the same Mac — the app refreshes the one it already has.
1.2.6 — Activation, made unrepeatable
Pressing Activate more than once could quietly register the same Mac several times, using up the licence’s device slots until the key refused the very Mac it was bought for. The app now recognises a Mac that is already activated with the same key and simply refreshes it — repeat presses cost nothing.
And when activation has already succeeded, a hiccup in the final screen refresh is no longer reported as a failure — the licence is on the Mac either way, and saying otherwise invited exactly those repeat presses.
1.2.5 — Subscribe in one click, and a launch-path fix
Voicecape Cloud can now be bought directly — the Subscribe button on this site and in the app opens a checkout where you pick monthly or yearly. Cancelling, switching plans and receipts all live in the customer portal, linked from the pricing page.
Fixed: if a development copy of the app had ever registered itself to start at login, that stale copy could keep launching — and failing to update — after the real app was installed. The app now refuses to register a non-release copy and quietly re-points the login entry at the copy you actually run.
1.2.4 — Licence housekeeping, done honestly
When our server definitively says a licence is finished — refunded, or the key was reissued — the app now retires its stored offline permission right away instead of letting it run out on its own. Nothing changes for a network hiccup or a server outage: those are not verdicts, and the app keeps working exactly as before.
The app also takes the chance to re-check its licence after a dictation ends, so a Mac that stays awake for weeks hears about changes without a restart. The check itself still leaves your Mac at most once every seven days — the same promise the privacy policy has always made.
1.2.3 — Quieter screen, smarter microphone
Settings gained a switch for the on-screen capsule. Turn it off for presentations and screen recordings — the sounds still play, so you always know a dictation started, finished or fell back.
If you switch apps while talking, the capsule now says so: ⇄ ⌘V means your words are on the clipboard because the app in front changed, not because the field refused them. Different fact, different message.
And when the microphone you chose is not connected — AirPods off, an interface unplugged — Voicecape now tries the microphones you used recently before falling back to the system default. Nothing to configure: your own picks are the priority list.
1.2.2 — Cloud dictation from the first launch
New installs now start with Cloud dictation on. Speech goes to our relay while you speak and comes back as text in real time — faster, and better at names, jargon and mixed-language sentences than the model inside the app.
If you already had Voicecape installed, nothing changed: your switch stays exactly where you left it. And the switch itself is still in Settings ▸ Processing — turn it off and recognition runs on this Mac only, with no audio leaving it.
One provider receives the audio, it is named in the privacy policy and listed in the Trust Center, and we keep no copy. When the network is unreachable the app falls back to on-device recognition rather than failing.
- The microphone permission prompt and the settings copy now describe what the switch does instead of asserting a default — in all five languages.
1.2.1 — Recognition in the cloud, if you want it
Until now the words became text on your Mac and there was no other option. There is one now, and it is off until you turn it on: Settings ▸ Processing has a Cloud dictation switch, and with it on the audio goes to a speech provider instead of to the model inside the app. It comes back faster and it is better at names, jargon and the languages that are hard for a small on-device model.
This is a different promise from the one the previous build made, so we will say it plainly: with this switch on, the audio leaves your Mac. Cloud polish only ever sent text. If that trade is not one you want, leave the switch alone — everything else in the app works exactly as it did, entirely on your Mac.
One provider receives it, it is named in the privacy policy and listed in the Trust Center, and we keep no copy. When it cannot be reached the app falls back to on-device recognition rather than failing.
- The default shortcut was doing nothing on most external keyboards. It is bound to Fn, and third-party keyboards handle that key in their own firmware and never send it — so the app was waiting for a key that could not arrive. Onboarding now says so and offers a shortcut that works, and if you already picked your own nothing changes.
- The microphone permission prompt said the app was going to do something it does not. Corrected in all five languages.
1.2.0 — Cloud polish, if you want it
Speech recognition has always run on your Mac and still does. What is new is an optional second step: after the words are already text, you can have that text cleaned up by a language model instead of by the rules that ship inside the app. It reads more like something a person wrote — full sentences, punctuation where it belongs, the false starts gone.
It is off until you turn it on. Settings ▸ Processing has a new option next to the ones that were already there, and if you never open that menu nothing about this build talks to a server that the previous one did not.
When it is on, the transcribed text — never the audio — goes to our relay and from there to one language-model provider. We keep no copy of it. The privacy policy names every provider and says exactly what the request carries, and the Trust Center lists the hosts.
It works during the trial too. The trial has always been the whole app, and this did not become the exception.
- Pick your microphone in Settings — the app used to take whatever macOS considered the default, which is wrong the moment you plug in an interface.
- The menu bar menu speaks all five languages, and so does the app menu next to the Apple logo.
- Recent dictations on the home screen, so the last thing you said is one click from being copied again.
- The shortcut picker no longer runs off the bottom of a small window.
1.1.1 — Buying it works
Until this build there was no way to turn a purchase into a working app. You can now paste the licence key from your receipt into Settings, press Activate, and that Mac is licensed.
The trial also became real: the app stops inserting text after 14 days, which is what the terms had been promising all along. Your settings and dictionary stay where they are and come back the moment a key goes in.
- One licence covers up to 3 Macs — stated the same way in the app, on the site and in the core, which it had not been.
- A Buy button next to the trial button, instead of a page that named a price and gave no way to pay it.
- Settings now says out loud that choosing your language beats automatic detection when you know what you will be speaking.
1.1.0 — Five languages, one model
English, Korean, Spanish, Japanese and Chinese, from a single model that ships inside the app — including sentences that switch language halfway through. The smaller model we had been shipping got Korean and English right and missed Spanish, Japanese and Chinese outright, so it was replaced.
That is why the download grew from about 215 MB to about 575 MB. It is a real cost and it buys the languages; nothing else in the app got heavier.
The second big change is that text stopped disappearing. Insertion used to trust the accessibility API when it reported success, and in browsers and Electron apps that report arrives before the write has actually happened — so the fallback never ran and the sentence was silently lost. Insertion is now judged by its effect instead of by a return value.
- Voicecape lives in the menu bar: closing the window no longer quits the product, and the Voice Bar appears while you speak.
- The personal dictionary is actually handed to the recogniser now — before this it was stored and never used.
- Automatic language detection, with a confidence gate so one unclear word does not switch languages mid-sentence.
- Punctuation is preserved, and line breaks come from real silence rather than from guesswork.
- The Settings window gained a sidebar; Dictionary and History moved inside it.
- fn is the default hold-to-talk key on macOS.
- Onboarding opens System Settings for you and moves to the next step when you come back.
1.0.1 — The first patch
A small release, aimed at the first thing everybody hit: onboarding could get stuck. Choosing a shortcut saved the value without re-reading what the app could now do, so the same step stayed on screen and there was no way forward.
- Shortcuts are shown the way macOS writes them — ⌘⇧Space rather than an internal name.
- The app and the core agreed on the default shortcut for the first time; before this they held different values and the one on screen was not the one that fired.