macOS and Windows · offline recognition and translation, no tokens
Say it. Read it.
Hold one key and speak; your words land at the cursor. Put ideas into everyday writing or Vibe Coding prompts, translate a region of your screen, and read web pages and video subtitles in two languages with optional local read-aloud.
Translation is fast and accurate, and all of it runs on your own computer — your audio and your words never leave it.
Install it once; four jobs, all on your own machine.
01
At your desk
Hold to talk, let go to write.
Hold right Command on a Mac, right Alt on Windows. Let go and the text lands at the cursor — chat box, office document, Vibe Coding prompt, or code comment, all the same. No switching windows, no copy and paste.
Mix Chinese and English for documents, chats and Vibe Coding prompts
Write by keystroke, by paste, or with a typewriter effect; long passages split into paragraphs on their own
Put proper nouns in the word list, or correct them outright with "wrong=right"
Use a system microphone or headset; clear audio and a quiet room help recognition
02
Reading the web
A whole foreign page, bilingual in one click.
The browser extension turns the page into side-by-side bilingual text, original next to translation, and one click puts it back. Select a phrase for a bubble, keep the floating button on the right within reach, and translate as you type in a separate window.
15 target languages, Traditional Chinese included
Body text only or the whole page — your call
The extension reaches Supersey over a local loopback port; pairing is a one-time step
03
Mockups and documents
Whatever you box, you can read.
Open the capture frame and drag it anywhere on screen — a design mockup, a chat window, a PDF that will not let you copy. The text inside is recognised and translated, and can sit over the original or off to one side.
Put the translation left, right, below, or over the original
Download translation models as needed, with clear notices for missing models or failed translations
Refreshes on the interval you set and retranslates when the picture changes
04
On the sofa
Subtitles translated as they play, and read out in step.
Subtitles on YouTube and Bilibili are translated into two languages while the video runs, auto-generated ones included. Press the speaker and a local voice reads the translation out loud in sync with the picture; the original audio drops in volume without going silent.
Works on any site that uses track subtitles
Chinese and English dubbing run locally; the first download is about 330–379 MB
Follows pauses and playback speed; long lines speed up without changing pitch
NEW · HUD GESTURES
Drag up to type. Drag down to capture.
Update and launch the app: gestures are already enabled. The colorful recording bar at the bottom of your screen — the HUD — now opens two translation tools.
Default recording keysWindows Right AltMac Right ⌘
Prefer Alt/Option on Mac? Choose Right Option (⌥) in Preferences → Voice Input → Hotkey, save, then restart as prompted. If you have customized your hotkey, use that key instead.
Drag upRelease to open
Supersey− ×
Original
会议下午三点开始。
Translation
The meeting starts at 3 pm.
↑ Drag up
Text translation
Open a floating window to type or paste text. It uses the same two-column layout as the browser’s right-click translator, with the original and translation side by side.
Drag downRelease to open
Supersey− ×
Original
会议下午三点开始。
Translation
The meeting starts at 3 pm.
↓ Drag down
Screen OCR translation
Open the familiar screen translator and select a region to read and translate. The existing window layout and region controls stay the same.
Illustrations · Use these gestures in the desktop app
01
Hold your hotkey
Wait for the colorful HUD at the bottom of the screen. No need to speak.
02
Click and drag the HUD
Hold the left mouse button on the HUD and drag up or down until the feature name appears.
03
Release at the prompt
Release the mouse when the open prompt appears. Drag back to the center before releasing to cancel.
↕
Already open? Repeat the same gesture to close it.
Bring up the HUD again, drag in the same direction, then release at the close prompt. Text drafts stay for the current app session; screen recognition stops. Clicking the HUD discards that recording without inserting text.
Activate the app and prepare the translation models when prompted. Screen OCR requires the Complete edition; Mac also needs Screen Recording permission. If the HUD is missing, check your recording hotkey and microphone access, plus Accessibility and Input Monitoring on Mac.
Why it stays local
Your voice and your words stay on your computer.
The message you have not sent, the proposal that is not final, the thought you jotted down. Recognition, translation, cleanup and dubbing all happen on the machine. No account, and no API key.
Recordings never hit the disk
Audio goes from memory straight to the local recogniser and is dropped once it is used. Nothing is uploaded.
Logs leave the text out by default
Everyday logs keep character counts and status only; logging the text itself is something you turn on yourself.
The clipboard is handed back
Before pasting it is backed up in every format and restored afterwards. Anything you copy in the meantime is kept.
Two platforms
The same hold-to-talk, with different strengths.
The core is identical: one hotkey model, the same three ways of writing text, the same screen and browser translation. The differences come from what each platform's hardware and system can do.
macOS
Apple Silicon
Switch recognition engines in Preferences and take whichever is faster on your machine
The recording overlay shows a live draft by default, so you read as you talk
Screen text recognition uses the system Vision framework, so it costs no extra space
Find settings with sidebar search, with multiple interface languages supported
Windows
Windows 10 or later · x64
Smart pause: the recording ends when you naturally stop, so you need not keep holding
Speech translation can switch to quality mode, which holds up better on long sentences
With an NVIDIA card detected, both recognition and translation can run on the GPU
Download models when needed and keep using the ones already installed
Getting started
Three steps to your first sentence.
1
Install the app
On a Mac, open the DMG and drag it into Applications. On Windows, run the installer, or unzip it and open Supersey.exe directly (keep the _internal folder next to it).
2
Follow the first-run guide
Activate and download the required speech model; translation and read-aloud models are optional. Allow Microphone, Accessibility and Input Monitoring on Mac, or microphone access for desktop apps on Windows.
3
Hold the hotkey and talk
Once the app is ready, focus a text field and hold right Command on Mac or right Alt on Windows. Right Ctrl starts voice translation by default; choose a left/right key or Ctrl + Shift combination in Settings.
One licence key binds to two devices, and you can unbind one yourself when you change machines. After activation it works offline for about four weeks, then needs one online check.
Yes. Audio goes from memory to the recogniser on your machine and is not uploaded. The first launch downloads a recognition model — about 2.5 GB on the Mac, about 1.02 GB for the quantised Windows build — and reuses the local cache after that. The translation and dubbing models work the same way.
How many languages are supported?
Recognition covers 30 languages and handles Chinese and English in one sentence. For translation, the browser extension and video subtitles cover 15 target languages, Traditional Chinese included; the speech translation hotkey and screen translation go between Chinese and English for now.
How does Traditional Chinese work?
Set the Chinese script to Traditional and every output switches to Taiwanese usage, not just the character forms: you get 軟體 rather than 软件, 設定 rather than 设置. The interface can be set to Traditional Chinese too, and it is available as a translation target.
Do I have to keep the hotkey held down?
Hold to talk and release to recognise is the default, but you can switch to press once to start and again to stop. Windows also has a smart pause mode that ends the recording when you naturally stop talking; that mode is not on the Mac yet.
Does the cleanup pass rewrite what I said?
It never changes your meaning. It applies rules only: spacing between Chinese and English, collapsing repeated punctuation, folding stutters like "this this", and fixing words that are commonly misheard. Both platforms use rules rather than a language model, and you can turn it off in settings.
How do I install the browser extension?
The extension connects to Supersey over a local loopback port. Confirm a four-digit pairing code in the app on first connection. Both Mac and Windows can export the Chrome extension from the Browser settings page; load it unpacked at chrome://extensions. Chrome 116 or later is required.
If a translation fails, do I lose the original?
No. If a speech translation model is not ready, times out or fails, the recognized original is kept and a notice appears. Screen and text translation show their current state or error; check the required models in Settings.
What about proper nouns it keeps getting wrong?
Add them to the word list in settings. A plain entry helps recognition; written as "wrong=right" it replaces the word before output. The app already ships with a batch of common terms and mishearing fixes, and your own entries take priority.
Does paste mode lose my clipboard?
No. The clipboard is backed up in every format before writing and restored afterwards. If you copy something new in the meantime, the app notices and keeps yours.
What if I change computers?
A licence key binds to two devices at most. Unbind the old one in the app settings, then activate on the new machine.
SUPERSEY · macOS & WINDOWS
Just say it.
The next message, the next note, the next page you cannot read.