Hold a key.Speak.It's typed.
Kazhy types into whatever app already had focus, so you never switch apps, open a transcription window, or paste by hand. On-device transcription runs entirely on your Mac: no account, no upload, and it keeps working offline.
A sentence this length, transcribed in the cloud, from the key coming up to the text standing in the field: about a second.
Sasha9:38
Staging is on the new schema.
Sasha9:40
The database is still catching up though.
at nikita→@nikitapostgres→Postgres Two replacements you wrote yourself, run after every dictation.
Where the words land
There is nothing to connect. Kazhy puts the text on the clipboard and presses ⌘V into the field that already has focus, so it works anywhere a text field does. These are examples, and the row keeps going because the list does.
- Google Chrome
- Gmail
- Notion
- Telegram
- Figma
- Linear
- GitHub
- Discord
- Jira
- Xcode
- Obsidian
- WordPress
Settings › Vocabulary
Two ways to fix a word.
One changes what the model hears. The other fixes what it wrote.
Words sent to the model Cloudflare
Replacements my email hello@kazhy.app
- Permissions
- General
- Transcription Engine
- Language
- Formatting
- Vocabulary
- History
- Recording
- Help
Vocabulary
Words sent to the model
Helps the model hear a word correctly. Space is limited.
Words01
Space0 / 1114 / 111
Space is what the model reads before a dictation. A longer word costs more, and a long list makes recognition worse, not better.
Replacements
Fixes what the model still writes wrong, after it writes it. No limit.
1 replacements · newest last
Re-format re-runs your rules. Re-transcribe re-runs the model.
opens the history window
Every field and button here works. Try one.
Send the release notes to [email protected] once Cloudflare is deployed.
64 symbols63 symbols70 symbols
Pushing the worker now, the D1 migration goes out with it.
58 symbols
Reverted the cache change, latency is back to normal.
53 symbols
AI Styles
The same dictation, four ways, one of them yours.
You choose which one runs. What it returns is what Kazhy pastes.
opens the AI Styles window
AI Styles
BUILT-IN
MY STYLES
- My styleCaveman
Prompt
3204 / 5000
✦Runs on Groq with your own key · nothing to set up
Duplicate to edit
Plain English, and the box holds 5,000 characters of it.
Caveman is not one of them. A custom style is one you write yourself, in plain English.
Every AI Style needs a Groq key, the six that ship and the ones you write.
Hey, so I was wondering if you could like maybe take a look at the you know the onboarding flow because people are dropping off at the permission step and I think it's because we ask for accessibility before we explain why we need it.
Hey, so I was wondering if you could like maybe take a look at the you know the onboarding flow because people are dropping off at the permission step and I think it's because we ask for accessibility before we explain why we need it.
Hey, so I was wondering if you could like maybe take a look at the you know the onboarding flow because people are dropping off at the permission step and I think it's because we ask for accessibility before we explain why we need it.
Hey, so I was wondering if you could like maybe take a look at the you know the onboarding flow because people are dropping off at the permission step and I think it's because we ask for accessibility before we explain why we need it.
Hey, so I was wondering if you could like maybe take a look at the you know the onboarding flow because people are dropping off at the permission step and I think it's because we ask for accessibility before we explain why we need it.
- Review the onboarding flow - Users are dropping off at the permission step - The drop‑off may be caused by asking for accessibility before explaining why it’s needed
Could you take a look at the onboarding flow because people are dropping off at the permission step and we ask for accessibility before explaining why we need it.
Hey, I was wondering if you could take a look at the onboarding flow because people are dropping off at the permission step, and I think it's because we ask for accessibility before we explain why we need it.
Ugh. Onboarding flow big maze. People fall leaf. Permission step big wolf chase. Ask accessibility before explain. Brain confused like sabertooth. Worker-man think cause. Bad.
↑↓ formatted/original
Six built-in styles ship with Kazhy. Every version above came back from a real run and was not written for this page.
Free and Cloud
Kazhy is free. The cloud runs on your own Groq key.
There is no subscription and nothing to sign into. Add a Groq key in Settings and transcription moves to the cloud and AI Styles turn on. Skip it, in onboarding or forever, and Kazhy keeps dictating on the model already on your Mac.
Local
$0 always
Your Mac does the work
- No key, no account, no setup
- Works fully offline
- Your vocabulary, your full history
- Nothing leaves your Mac
✦Cloud
$0 on Groq’s free tier
Paste a Groq API key in Settings, or skip it during onboarding
Faster, and unlocks AI Styles
- Your own Groq key, not ours
- Audio goes straight from your Mac to Groq
- All six AI Styles, and any you write
- A key is free to create at console.groq.com
A Groq key costs nothing to create, at console.groq.com/keys. Kazhy never sees it before you paste it in, and it goes straight from your Mac to Groq from then on.
Both lists above describe the app as it actually runs, not a plan you choose once.
Questions
Before you download.
Does the cloud path cost anything?
No. Kazhy is free, full stop, and so is a Groq account. Groq's free tier covers ordinary dictation with room to spare, and if you ever do hit its rate limit Kazhy just finishes the job on your Mac instead of failing.
What happens if I never add a key?
Kazhy keeps working. Dictation runs on the model already on your Mac, fully offline, with no account and no daily limit. A key only changes two things: it moves transcription to the cloud, and it turns on AI Styles.
Where does my audio go?
With no key, nowhere. It never leaves your Mac. With a key, it goes straight from your Mac to Groq, on your own Groq account, and Kazhy has no server of its own in between. See the privacy policy for what that means in practice.
Can I write my own AI Styles?
Yes, once a key is connected. A style is a name, an emoji and a prompt in your own words, plus a switch for whether it may rewrite what you said or only tidy it. The six that ship cannot be edited, but any of them can be duplicated and the copy changed.
Which languages can I dictate in?
99. You choose the one you are speaking in — in Settings › Language, or on the recording orb, which moves through the languages you keep. Kazhy transcribes what you said; it does not translate between languages.
Why is it not in the Mac App Store?
Kazhy types into other applications and listens for a global hotkey, which is deeper access than App Store sandboxing allows. So it ships directly instead: signed with a Developer ID, notarized by Apple, and updated in place.
What does it run on?
macOS 14 or newer, on Apple Silicon — M1 and up. It asks for Microphone permission to hear you and Accessibility permission to type into the application you are already in.
Say it once. Read it back where you were already working.
Download for MacFree forever on your own Mac, with no account. Add your own Groq key for the cloud and AI Styles, whenever you want to.
macOS 14 or newer, Apple Silicon. Microphone and Accessibility permissions.
Kazhy for macOS. Every Kazhy window on this page is the real application, redrawn for the web.