Dictation
⌥S
Hold ⌥S and talk. It’s typed into any app, with the ums taken out.
See it work →
YAPYAPOn-device AI productivity
Own your AI.
AI that runs entirely on your Mac, not on someone else’s servers.
Free · Open source · macOS 15+ · Apple silicon
The roadmap
What you can do today.
Already replaces Wispr Flow and Speechify. Tomorrow, who knows?
Read aloud
⌥A
Press ⌥A and draw a box around any text. You hear it, word by word, and catch what your eyes skip.
See it work →Spoken shortcuts
⌥S
Save your email, address or API keys. Say the name, and it’s pasted.
See it work →- NextUnderstand your filesAsk about your documents without uploading them.
- LaterOrganize your workSort, tidy and find things for you.
- LaterAct across your MacAn assistant that gets real tasks done.
Each step tests how much your own computer can do, until it can do what the cloud does.
Demos
See it work.
Speak anywhere. Hold ⌥S and talk. Every example is a real transcript.
Messages
Reply without typing.
# engineering
D
Dana
Login is broken for some people on staging.
Team chat
Post an update without leaving your code.
Letter to Dr. Lee
Dear Dr. Lee,
Documents
Write letters and reports out loud.
Reads every app. Press ⌥A and draw a box around any text. You hear it word by word.
PSPriya ShahThursday offsiteTo: Design Team
Hi all, the offsite has moved to the fourth floor. Bring a laptop and your third quarter numbers. Lunch is at noon, and we should wrap up by four.
Listen while you do something else.
recall-study.pdfPage 1 of 8
Sleep and Recall in Adult Learners
Adults who slept within three hours of a study session recalled more of the material a week later. The effect was largest for new vocabulary. Participants who stayed awake showed no gain from extra review time. We discuss what this means for evening classes and for learners who work night shifts.
PDFs
Columns read in the right order.
IMG_204112 Oct 2026
4.2 PhotosynthesisPlants use energy from light to turn water and carbon dioxide into sugar. Oxygen is released as a by-product.
Photos
Even text inside images.
Say the name. It’s pasted. Save your email, address or API keys, then hold ⌥S and say its name.
New Message
To:
Subject:Saturday
Count me in. Send the details to the address above.
“Type email address”
Fill in forms without spelling it out.
shop.example
Shipping address
Full nameSam Rivera
Street address
“Home address”
Check out in seconds.
yapyap — zsh
~/yapyap % export ELEVENLABS_API_KEY=""
“Paste my ElevenLabs key”
Kept in your Keychain, never in history. Touch ID if you want it.
Why I’m building YapYap.
Hi, I’m Asad. I think the AI you use every day should run on the computer you own, not on someone else’s servers. YapYap is me finding out how far a Mac can go on its own: dictation and read-aloud first, then your files, then an assistant that gets real work done.
To explore ideas, and because I’ve always wanted a TV show, I made a little TV show about this mission. It’s fun. We rarely talk about the product; we’re usually goofing around. Find us on Instagram and TikTok.
BobSupport LeadTypos are just the email being creative.
WaltFounder and CEOI’m not mad. I’m just printing this out so I can be mad on paper.
ZoeSummer internHold on, I’ll just say it to you. Um, like, the meeting is moved. Sent.
LennyIT and SecurityI unplugged the printer. It knew too much.
GailOffice ManagerWe had a good run. Four minutes.
JessyHead of SalesMute button? I don’t believe in it.
LENNY · IT AND SECURITYBefore I say yes: does it run locally?
Privacy
What you read and say never leaves your Mac.
Yes, Lenny, it runs locally. You don’t have to take our word for it, though. Here is what the code does, and how to check.
One-time downloads.
The natural Kokoro voices, about 340 MB from GitHub, and NVIDIA’s Parakeet speech model, about 700 MB from Hugging Face, both fetched in the background the first time you open YapYap (you can remove them in Settings). And on macOS 26 or later, Apple’s speech model, which macOS fetches itself. Text, audio and screenshots are never uploaded.
Kokoro.swiftScreenshots stay in memory.
Each read takes one still image of the window or box you chose. macOS’s Vision framework turns it into text, then the image is thrown away. When you dictate, YapYap does the same with the window you’re typing in, only to spell the names in it right (you can turn this off).
Capture.swiftYour voice stays on your Mac.
Dictation runs NVIDIA’s Parakeet speech model on your Mac (or Apple’s on-device speech recognition, if you prefer it or until Parakeet has downloaded), and the optional tidy-up uses Apple Intelligence on your Mac. The mic is on only while you dictate: while you hold the shortcut, or in hands-free mode until you stop it. Audio is never saved.
Parakeet.swiftNo account. Anonymous stats you can switch off.
YapYap sends anonymous counts, like how many words were read, through TelemetryDeck. Never what you read or say, or which app it was in. Turn it off in Settings → Privacy.
Telemetry.swiftCheck it yourself.
- Search the code. YapYap’s own network code fits on three lines; the open-source FluidAudio library fetches the speech model from Hugging Face. The only other connections are TelemetryDeck’s anonymous stats and a daily check of yapyap.cc for updates.
- About this website. yapyap.cc itself counts visits, clicks and anonymous replays of how the page is used with PostHog, without cookies and through yapyap.cc. That’s the website only: the app never talks to PostHog.
- Turn off Wi‑Fi. Once the voices and speech models are installed, reading and dictation work exactly the same.
- Watch the traffic. Run a network monitor such as LuLu while YapYap reads or listens. Turn stats off and the only other request is the daily update check.
$ grep -rnE 'URLSession\(|https://' Sources AppModel.swift:476 session = URLSession(configuration: …) Kokoro.swift:253 "https://media.githubusercontent.com/…/kokoro-v1_0.safetensors" Kokoro.swift:254 "https://raw.githubusercontent.com/…/voices.npz"
Open source
Free, and every line is on GitHub.
Read it, build it yourself, or help make it better. Licensed under the AGPL, so it stays free.
- Swift & SwiftUI
- Vision
- Speech
- MLX
- Kokoro-82M
- Parakeet
$ git clone https://github.com/WannaBeSolopreneur/ YapYap $ cd YapYap $ scripts/build-app.sh --install
FAQ
Questions and answers.
GAIL · OFFICE MANAGERI put the sign up. Nobody reads signs. I should know, I work here.
What is YapYap?
YapYap is a free, open-source Mac app that does two things. It reads any text on your screen aloud, highlighting each word as it’s spoken, and it types what you say into any app. Reading and dictation both run on your Mac, with no account.
How do I get my Mac to read text aloud with YapYap?
Press ⌥A, then drag a box around the text you want to hear. YapYap finds the words, even in screenshots, PDFs and photos, and reads them aloud while a highlight follows each word on the original page. Press Esc to stop.
How does dictation work?
Hold ⌥S and talk. Your words appear live in a small panel at the bottom of the screen. Let go and YapYap types the text wherever your cursor is, in any app. For longer dictation, tap ⌥S instead: it keeps listening until you tap it again or click ✓. It can drop filler words and tidy the text with Apple Intelligence, on your Mac. Save things you type often under Saved, like your email address or an API key, and dictate just the name to paste them.
Does YapYap send my voice or screen anywhere?
No. Text on screen is recognised with Apple’s Vision framework, dictation runs NVIDIA’s Parakeet model on your Mac, and the voices run locally. YapYap only sends anonymous usage counts, never content, and you can turn that off in Settings. (This website, separately, uses cookieless PostHog analytics to see how the page is used.)
Does YapYap work offline?
Yes. Once the Kokoro voices and the speech models are downloaded, reading and dictation both work with Wi‑Fi turned off. Until the voices arrive, or if you remove them, YapYap reads with the voices built into macOS.
Which Macs does YapYap run on?
YapYap needs macOS 15 Sequoia or later on a Mac with Apple silicon (M1 or newer). It asks for Screen Recording permission to read, and Microphone and Accessibility permission to dictate.
What languages does YapYap support?
The natural Kokoro voices speak US and UK English. Dictation uses NVIDIA’s Parakeet for English and 24 other European languages, and Apple’s on-device speech recognition for other languages Apple supports. YapYap is designed and tested in English.
How is YapYap different from Speechify or Wispr Flow?
YapYap is free and open source, with no subscription. It handles both reading aloud and dictation in one app, and it does both on your Mac instead of sending your text or voice to a server.
How do I install YapYap?
Download YapYap, open the file, and drag YapYap onto Applications. It’s signed and notarized by Apple, and it keeps itself up to date. You can also build it from the source on GitHub.
