New in 1.3Live translation between any two of sixteen languagesLive translation, on device

Live captions for anything your Mac plays.

Calls, videos, podcasts and lectures, transcribed as they play and drawn in an overlay that stays above every window. Translated live if you want it. Nothing leaves the Mac.

  • Free for 7 days, no card, no account
  • $9 once, every update included
  • 14-day refund, no questions asked
  • Download for macOS 14.2 or later, Apple silicon
  • Rated 5.0 on Gumroad

Switch apps and the captions stay: the overlay is above every window, and lets clicks through.

Click a window to bring it forward. Hold and drag the captions to move them, as you would in the app.

Point at the captions and they dissolve under you. Hold to stack the last few back up. Both work here exactly as they do in the app.

Use cases

Built for the moment you lose the thread

In a meeting, second language or not, one missed word costs you the next three. Subtitles captions whatever your Mac is playing, in the app it is playing in, so you keep up without asking anyone to repeat themselves. Whether you are deaf or hard of hearing, hearing but not following, listening in a second language, or just working somewhere loud, three situations come up again and again.

02

Working while you listen

Switch to your notes, your editor or a browser. The overlay stays above every window and lets clicks through, so the conversation carries on without you watching it. A notepad like Granola writes the meeting up afterwards; this is the half that has to happen while it is still going on.

03

Videos, podcasts and lectures

A video with no captions, or with automatic ones that are worse than none. A podcast in a language you're still learning. A recorded or a streaming lecture, an interview, a film with quiet dialogue. Anything that plays is captioned as it plays, in its own language or translated into yours.

Privacy

Nothing leaves the Mac. Nothing for IT to review.

The reason you can point this at a work call is that it is not a service. There is no server, no account and no transcript, so there is nothing to send and nobody else in the loop. The long version is written down. The source is published, so none of it has to be taken on trust.

  • No bot joins the call. It captures the audio your Mac is already playing, so nothing appears in the participant list and nobody is told anything is running. Most of this category sends a visible participant instead, which plenty of companies disallow outright.
  • Nothing is recorded, and nothing is written to disk. Audio goes from the tap into the recogniser and is discarded on the spot. Captions are on screen and then they are gone; the boxes brings back are held in memory until you quit, and there is no file to find afterwards, on your Mac or anywhere else.
  • The microphone is never opened. It only ever hears what your Mac plays. That is a real limit, and it is also why nothing you say can go anywhere.
  • No account, no telemetry, no network. Nothing to sign into and no usage reported anywhere. The model downloads once, about 633 MB, and after that the only request left is a daily check for a newer version, which it asks you about first: pull the Wi-Fi and it keeps captioning.

Features

Built around three constraints

Live audio can't be paused, can't be re-run, and shouldn't have to leave your machine. Everything here follows from that.

Nothing leaves the Mac

It runs locally and offline: no accounts, no telemetry, no network call carrying anything of yours. Audio goes from the tap straight into the recogniser and is discarded. The only things ever downloaded are the model, once, and an update when you choose one.

Neural Engine, not CPU

Roughly 0.15 real-time factor: six seconds of speech transcribed per second of compute. The same model on the CPU measured about 100× too slow to keep up with a live stream.

Every app, or just one

It listens to everything your Mac plays by default, so there is nothing to set before it works. Narrow it to a single app from the menu bar when two things are talking at once, and it captures that app plus every helper process it spawns, which is what makes browsers and Electron apps work, where audio never comes from the main process.

Music doesn't become lyrics

A voice detector runs ahead of the recogniser so backing tracks never reach it. Without it, music fills the model's context and the first words after it are lost.

A new box per speaker

Optionally start a fresh subtitle when someone else starts talking, the same way a pause does. Off by default: it runs a second model alongside the first.

Sixteen languages

The default model detects the language itself. English gets seven dedicated checkpoints on top, trading latency against accuracy from 160 ms upward.

A hole where you point

Move the pointer near the box and it fades away underneath it, so the sentence you are reading never covers the sentence you are writing. How much of it goes and how wide the hole reaches are both yours to set. Hold to keep it solid.

Hold ⌥ for what you missed

The box pages like broadcast subtitles: it fills, clears and starts again, so a line you glanced away from is gone. Hold and the finished boxes stack back up above the live one, every one of them since the last pause in the talk, scrollable for the older ones and searchable with F. Let go and they are gone again. Nothing is written anywhere.

The overlay is the preview

, from the menu opens settings: how many lines a box fills before it clears, how solid it is, how far the pointer reveal reaches, how far back goes. Every slider applies to the overlay as you drag it, so you set these by watching them rather than by picking a number.

How it works

From the app's audio to your screen, in four steps

A realtime audio thread that only ever copies, a portable core that never transcribes, and a model running where it belongs.

  1. 1

    Core Audio process tap

    48 kHz stereo, a callback every 10.7 ms. The realtime thread copies into a lock-free ring buffer and does nothing else.

  2. 2

    Resample and gate

    Down to 16 kHz mono on a worker thread, silence skipped, with about a second replayed ahead of speech so no word starts cold.

  3. 3

    Parakeet on the Neural Engine

    A streaming model, not a 30-second-window one: the difference between captions that track the audio and captions that arrive seconds late.

  4. 4

    Overlay

    Three-line pages that clear and restart the way broadcast subtitles do, fading four seconds after the last new text rather than after the audio stops.

Everything from the ring buffer to the model boundary is portable Rust that deliberately doesn't transcribe: it hands out frames. The tap and the overlay are the only Mac-specific parts.

Compare

Why not Apple's Live Captions?

macOS has captions built in, free, on every Apple silicon Mac, under System Settings, Accessibility, Live Captions. If that covers what you need, use it. These are the places it stopped covering mine, and there is a row-by-row comparison if you want the long version.

Feature Subtitles Apple Live Captions
Languages Sixteen, detected automatically, translated live between any pair Not available in every language or region
The overlay Click-through, fades under the pointer, stays above every window A window you move and resize
What it listens to Everything the Mac plays, or one app of your choosing Everything the Mac plays
Music Skipped by a voice detector, so a backing track never becomes lyrics Captioned
Speaker changes Optional new box per speaker No equivalent
The model Seven English checkpoints trading latency against accuracy, plus the multilingual one One, fixed
Price Free for 7 days, then $9 once, every update included Free, built in

What people say

Used on real calls

So smooth, very useful during meetings on different apps (slack, zoom, google), so well integrated on the screen, doesn't block interacting with the other windows, love it!
Remi Santos Verified purchase on Gumroad
Almost no setup, no tweaking around: you install it and it just works. Simple, useful and really effective.
Jérôme Blondon Verified purchase on Gumroad
Very handy! I love how it can be connected to any audio source on my laptop, which helps not only for meetings and videocalls, but also for all those streaming services that do not support subtitles, like news websites or even some YouTube videos. The model supports many languages, so it's also useful if you're learning a foreign language and need subtitles to help you understand what's being said.
Bruno Verified purchase on Gumroad
5.0

Average rating on Gumroad

14-day refund, no questions asked

Buy for $9

Why this exists

I work at an American company and my meetings are in English, which is not my first language and not most of my colleagues’ either. I would be halfway through something else when the part that mattered was said, and it would be gone: a word I did not catch, an accent I was not used to, two people talking over each other. Asking for it again costs everyone thirty seconds and interrupts whoever was speaking.

So the captions had to stay readable while I was doing something else. That is why they are an overlay above every window rather than a panel inside one app, why they take no clicks, and why they fade under the pointer instead of needing to be moved. Everything else on this page came after that.

Mathieu Jouhet, who makes it.

Pricing

One price. Nothing recurring.

Try it free for 7 days first, with no card and no account. Then $9, once: no minutes to buy, no seats, no renewal, every future update included, for less than one month of any transcription subscription.

  • Checkout by Gumroad. VAT, receipts and refunds are handled there, and the key arrives by email.
  • Signed and notarised, so the permission you grant survives every update.
  • Source published. Read it, or build it yourself.

Subtitles for Mac

$9 one-time

  • Free for 7 days first, no key needed
  • Unlimited captions, for as long as you like
  • Every future update included
  • Sixteen languages and live translation
  • One key for every Mac you use. No account, no subscription, no telemetry
  • 14-day refund, no questions asked

One-time $9 · macOS 14.2 or later · Apple silicon
About 633 MB downloaded on first launch.

FAQ

Questions people ask first

Does it work with Zoom, Teams and Google Meet?
Yes. It taps the audio an app plays rather than integrating with any one service, so a Zoom window, a Slack huddle and a call in a browser tab all caption the same way. Browsers and Electron apps included: it captures the helper processes those spawn, which is where their audio actually comes from.
Can it translate what it hears?
Yes, since 1.3. Translate To in the menu takes any of the sixteen languages into any of the other fifteen, through Apple's on-device translator, so a translated caption leaves the machine no more than an untranslated one does. It is off until you pick a target, and holding shows the original language underneath while it runs. It needs macOS 15; captioning alone still runs on 14.2.
Doesn't an overlay get in the way?
It takes no clicks at all, so you work straight through it, and it fades away under the pointer so you can read what is beneath it. How much it fades, how far that reaches, how solid the box is and how many lines it holds are all settings. Hold to keep it solid and drag it somewhere else; where you put it is remembered.
Does it caption my own voice?
No. It only ever hears what your Mac plays, never the microphone. On a call that means you read the other people, not yourself.
Does it need an internet connection?
Once, to fetch the model on first run, about 633 MB. After that it runs entirely offline. The only other request it can make is a daily check for a newer version, which it asks permission for first and which you can decline.
Can I save or export a transcript?
No, and that is deliberate. Nothing is written to disk: captions exist on screen and then they are gone. Holding brings the finished boxes back while the app is running, and F searches them, which is for finding the line you looked away from, not for keeping a record: they live in memory, a long enough silence clears them, and quitting ends them. If you need a record of a conversation, this is the wrong tool.
Does IT have to approve it?
There is no vendor here to approve: no account, no server, no transcript leaving the Mac, and nothing joining the call as a participant, so nobody in the meeting is being recorded by anything. It is one signed, notarised app on one machine, and after the model downloads once the only request it can make is a version check you can switch off. Whether that clears your own policy is your call, but there is no third party in it to review.
Does it work with headphones or AirPods?
Yes. It captures the audio an app produces, not the sound coming out of a speaker, so what you listen on makes no difference.
Will it slow my Mac down?
Transcription runs on the Neural Engine rather than the CPU, at roughly a 0.15 real-time factor, or six seconds of speech per second of compute. The same model on the CPU measured about 100× too slow to keep up at all.
Which Macs does it run on?
Apple silicon, macOS 14.2 or later. The version floor is Core Audio process taps, which is the API the whole thing is built on.
Is it real-time?
Yes. The model is a streaming one, so a caption appears while the sentence is still being said rather than after it ends. It reads the system audio an app plays, with no loopback driver to install, and the English checkpoints trade latency against accuracy from 160 ms upward. If you searched for speech to text or real-time transcription for the Mac, this is that, drawn on screen rather than saved to a file.
Is it a subscription?
No. The download is a free seven-day trial. A licence key is $9 once, every future update included, no account to create and no minutes to buy; it covers every Mac you use, and the app checks it with Gumroad about once a month, sending the key and nothing about you. If it is not for you, there is a 14-day refund.
Can I see the source?
Yes. It is published under FSL-1.1-ALv2, which becomes Apache 2.0 two years after each release. Read it, or build it yourself, at github.com/daformat/subtitles.

Captions for anything, without sending audio anywhere.

14-day refund, no questions asked

Free for 7 days, no card, then $9 once, every future update included · macOS 14.2 or later · Apple silicon
Unlimited live captions, for as long as you own the Mac. No minutes to buy, no account, no subscription. Read the setup guide.