ON-DEVICE VOICE TYPING / WINDOWS + MAC / ONE LICENCE
You talk. It types.
It even presses Enter.
Count how many replies you type today. Now count the Enters. Say your message and it appears in any app. Say "send it send it" and it's sent. Said the wrong thing? "clear clear" takes it back. Your hands never return to the keyboard — and your voice and text never leave your PC.
Buy once at the $99 launch price — no subscription, no account. Try the live demo above first.
Type a sentence, then run “send it send it” — or tap 🎤 and say it. Real usage is voice; you say the command twice so it never fires by accident. (The 🎤 here borrows your browser's speech engine, which sends audio to the cloud — a taste of the interaction. The real Kotocast recognizes everything on your PC: nothing leaves it.)
voice-input --status
Sound familiar?
Type, Enter. Type, Enter. Forty times a day.
Every reply pulls your hands back to the keyboard. Even with dictation, you still reach over to press Enter — and somehow that one keystroke is the most annoying part.
Messages arrive exactly when your hands are busy
Cooking. Flipping through papers. Holding a coffee. "I'll reply later" piles up — and you become the slow one in every chat.
Cloud dictation ships every word to someone else's server
Convenient — but that note to your family and that work discussion travel to some stranger's server. Your voice is far more personal than your typing.
kotocast --features
With Kotocast
Types into whatever app you're using
Chat, email, documents, terminals. Your words go straight into the focused window. No plugin, no per-app setup.
Finish the whole reply by voice
Start, send, clear, undo, copy, paste, screenshot, OCR, record — 25+ commands, all spoken. From first word to sent message, your hands never touch a key.
100% on your machine
Speech recognition runs entirely on your PC. Your voice and text never leave it. One model download on first run, and that is all it needs. (The Mac build checks your licence once a week — licence key and device ID only.)
Keeps up with your speech
On a supported GPU, each utterance lands in 0.2–0.35 seconds. You write at the speed you talk.
voice spells ── say it twice and it becomes a command. Said once, in a normal sentence, it never fires.
Plus select all, tab, redo, OCR, screenshot, record ── 25+ commands. All voice, zero keys.
→ Privacy details — exactly what connects, and what never does
kotocast --why-it-never-misfires
Voice-powered Send, without the misfires
This tool takes one sentence to explain: you talk, it types, your voice can send. Everything hard came after that sentence — the slow, unglamorous work of making sure it never fires when it shouldn't.
Why the command is "send it send it" — not "send send"
We wanted "send send". Speech recognition kept hearing "and send". "on on" vanished entirely. So every candidate phrase was played through multiple synthetic voices, over and over, and only the words that never broke made the cut. These aren't the phrases we liked. They're the phrases that survived.
Say "please send this to her" — nothing happens
A command only exists as a doubled phrase like "send it send it" — a shape that never occurs in normal speech. Dictate a message about sending, clearing, or copying all day long. Your words and your commands cannot collide.
Send has a 0.5-second grace window
The one action you can't take back doesn't execute instantly. Say "clear clear" inside the window and the send is canceled. A slip of the tongue never becomes a message you regret.
When you say nothing, it types nothing
Speech models famously hallucinate phrases like "thanks for watching" out of pure silence. Kotocast double-checks the model's own confidence and throws those ghosts away — so room noise and long pauses never end up in your chat.
kotocast --before-you-buy
Four questions to ask any dictation tool
Ask them of anything you're considering — including this. The answers decide more than a feature list does.
| The question | KOTOCAST'S ANSWER |
|---|---|
| Where is my voice actually processed? | On this PC. Nothing is transmitted — there is no server to send it to. |
| What happens the month I stop paying? | Nothing happens. You paid once; it keeps working. |
| Does it still work with the network unplugged? | Yes. After the one-time model download, it never needs the internet again. |
| Do I need an account to use what I bought? | No. A licence key, and that's it. |
# We deliberately don't print other products' prices here. They change, and a stale comparison table is just a false one. Ask them the four questions yourself — those are the ones that decide how the tool feels a year from now.
git log --author=maker
Who's building this, and why
No celebrity endorsements. No fake install counts. Just one engineer who ran out of patience with his own keyboard.
The engineer other engineers call when things break
At work, the gnarly problems land on his desk. Then one daily annoyance — type, Enter, type, Enter — finally got under his skin. This isn't a weekend hack. It's everything he knows, aimed at one stupid keystroke.
The boring parts got the most hours
Phrase-survival sweeps, doubled commands, grace windows, hallucination filters. The hours went into reliability nobody sees — which tells you more about the builder than any bio could.
He writes every email with it
Chat, email, work tools — voice, all day, every day. If it isn't good enough for his own inbox, it doesn't ship.
Built in the open
The whole process is public. Early-access users steer the roadmap — your "I wish it did X" becomes the next release.
kotocast --who-its-for
Sit back. Or never reach at all.
Give your posture a break. Hunching over a keyboard quietly wears the body down. Lean back, rest your hands, keep working — your voice does the reaching. Some people who tried it say composing out loud even sharpened how they order their thoughts.
When hands aren't an option. RSI, injury, illness, or plain fatigue — the keyboard is out of reach for all kinds of reasons. Kotocast lets you write and control everything by voice alone, so the experience is the same whatever the state of your hands. For people who want to, and for people who need to.
kotocast usage --meter
It doesn't get more expensive the more you use it.
No tokens. No monthly bill. No sign-in. You pay once — and your ten-thousandth sentence works exactly like your first. Your voice never leaves this PC. It keeps working with the network unplugged. (On Mac, a weekly licence check; go two weeks without one and it pauses — so a flight or a dead zone is never a problem.)
kotocast demo --uncut
There is no trial. So watch all of it first.
Eight demos, in the order they were filmed, at real speed. Nothing is cut and nothing is sped up — the moments where it mishears me are still in there. The small window in the corner is my actual hands.
Nothing reaches YouTube until you press play.
kotocast license --pricing
Buy it once. Own it.
Built for people who are done renting software. This is the launch price — it goes up to the regular price once a set number of copies have sold.
One-time purchase — Windows + Mac
$99 launch price
- Voice writing + 25+ spoken commands + custom spells + English, Japanese, French, German & Spanish
- ★ Area OCR — pull text out of images, PDFs and error dialogs to your cursor
- ★ Region screenshot — draw a box, it's on your clipboard (frame stays visible)
- ★ Screen recording of any region — start & stop by voice
- Yours forever + 1 year of updates included · major upgrades always optional
- One-time activation — your voice and text aren't sent even then (Mac: a weekly licence check, key and device ID only)
- ★ Both builds — Windows (.exe) and macOS (.dmg) — one licence, up to 2 machines
One key. Two desks.
Buying it once gets you both builds. There is nothing to buy again.
Kotocast for Windows — .exe
- Windows 10 / 11 (64-bit, x64). Windows on ARM is not supported
- With an NVIDIA GPU, 0.2–0.35s per utterance (measured on an RTX 4070). Runs on CPU without one
- Activates once, and never checks again at startup
Kotocast for macOS — .dmg
- macOS 26 or newer, Apple silicon (M1 and later). Intel Macs are not supported
- Notarised by Apple — download it and it opens
- Checks your licence once a week (licence key and device ID only)
One licence covers up to two machines. The Windows box at work, the Mac at home.
Same key, same voice, same behaviour.
No feature held back — it's all in one price. A subscription is a bill that arrives every year, for as long as you use it. This one arrives once.
kotocast --check
What it needs to run
Modest requirements, and a built-in --check tells you if your machine is ready before you commit.
- OS — Windows
- Windows 10 / 11 (64-bit, x64) — Intel / AMD. Windows on ARM (Snapdragon, etc.) is not supported.
- OS — Mac
- NEWmacOS 26 or newer, Apple silicon (M1 and later). Intel Macs are not supported. On macOS 14 / 15? Get in touch.
- Processor
- Any modern CPU — no GPU required to run
- GPU (optional)
- An NVIDIA GPU makes it fast — 0.2–0.35s per utterance (measured on an RTX 4070). Without one it still runs on CPU, just slower.
- Disk
- ~1.5 GB one-time voice-model download — nothing more after that
- Internet
- Not required to run. Only the first-run model download touches the network. The Mac build also checks your licence once a week (licence key and device ID only — no audio, no text).
- Languages
- English (default) + native-grade Japanese, French, German, Spanish, switchable in settings
- Setup
- Download and run — the built-in
--checkself-diagnoses your machine in one step - License
- Buy once — no subscription, no account, yours to keep (Mac: a weekly licence check)
- Both
- INCLUDEDOne licence covers Windows and Mac (up to two machines). Buying it hands you both the
.exeand the.dmg.
man kotocast | grep -A2 "Q\."
Questions, answered
Does my voice really never go to the cloud?
Really. Recognition runs on AI models stored on your PC. Your voice, your text and your usage data are never transmitted. The only network traffic is the one-time model download on first run — plus, on the Mac build, a weekly licence check that sends the licence key and a device ID and nothing else. (Full details in the PRIVACY doc.)
Do I need a GPU?
A GPU (NVIDIA and similar) makes it fast — 0.2–0.35s per utterance. Without one it runs on CPU, just slower.
Which apps does it work with?
Whatever app has focus, on either OS. Chat, email, browsers, documents, editors, terminals — nothing to install on the app side.
Does it work without internet?
Yes. After the first-run model download, dictation keeps working on a plane, off the grid, or inside a locked-down secure network. On the Mac build the licence is re-checked weekly; if it can't be checked for two weeks the app pauses, so two weeks off the network is fine. The Windows build never re-checks.
What languages are supported?
English, Japanese, French, German and Spanish are supported and switchable in settings. More languages are in the works, gated by the same phrase-survival testing.
Which OS?
Both. Windows 10 / 11 (64-bit) and macOS 26 or newer on Apple silicon. One licence covers both, up to two machines — buying it hands you the .exe and the .dmg together. Intel Macs are not supported; on macOS 14 / 15, get in touch.
Windows says "protected your PC" when I install — is it safe?
Yes. As a solo developer I haven't bought a code-signing certificate yet ($300–500/yr), so Windows SmartScreen shows its standard warning for unsigned apps — it does not mean the app is dangerous. Click More info → Run anyway to launch it. Everything runs locally; your voice and text are never transmitted. The built-in --check self-diagnoses your machine before you commit. A signing certificate is planned as adoption grows.
Will macOS refuse to open it?
No. The Mac build is notarised by Apple. Download it and it opens — no right-click, no Gatekeeper detour.
How do I request a feature or report a bug?
Please do — it reaches the developer directly, via the Xinogi site. This is built by one person, and I want to make it better together with the people who use it. "I wish it did X" and "this should work like Y" are exactly what I want to hear.
kotocast --buy
Buy once. Then forget we exist.
No subscription, no account, no cloud. Pay once at the $99 launch price (it goes up to $139 once a set number of copies have sold), download, and it's yours — running entirely on your machine. One licence, Windows and Mac both.