Free
$0
2,000 words a week. Cleanup, rewriting and vocabulary all included. No card, no account.
DownloadTalk. It types.
Takumi Voice is one key. Hold it, say what you mean, and finished text lands where your cursor already is. No window, nothing to clean up.
Rin
Are we still aiming for this week?
Transcribing
Inserted
Error Transcription timed out
Holding the key and saying “Um so I was thinking uh we should we should ship the the new pricing page on Thursday you know and then um send the announcement email right after.” types “I was thinking we should ship the new pricing page on Thursday and then send the announcement email right after.” into the app you were already in.
2,000 words a week free, forever
No account no API key, no setup
Mac, Windows, Linux three machines, one licence
Nothing running when you're not talking
Your keyboard is slow
Nobody thinks at forty words a minute. That's just how fast hands go, and it's been the ceiling on writing for as long as there have been keyboards.
Typing 40 wpm
Average, sustained, on a full keyboard
Speaking 150 wpm
Normal talking. No rushing.
Typical human rates, not our benchmarks. Yours will be different and the gap won't be by much, which is why every dictation tool quotes something like it. The gap only counts if you don't spend it again fixing the output. That's the part we build.
How it works
One gesture. Nothing to open, nothing to close afterwards.
- 01
Hold the key
Pick it by pressing it. A modifier you never use on its own works best — one finger, and it does nothing by itself.
00:02TranscribingInsertedError Transcription timed out - 02
Say it
A small bar shows it's listening and how long you've been talking. It never takes focus, so you stay where you were.
00:06TranscribingInsertedError Transcription timed out - 03
Let go
The ums come out. The punctuation goes in. You get the sentence you meant, not the one you said.
00:02CleaningInsertedError Transcription timed out - 04
It's typed
Text appears where your cursor already was. We borrow your clipboard for a second and put it back.
00:02TranscribingInsertedError Transcription timed out
That bar is the whole interface. It's the only thing we put on your screen, and it's gone the second your text lands.
It's not a transcript
Every pair below came out of the shipping pipeline, pasted here unedited — messy half included. Same gesture, different shape depending on where your cursor is.
You said
Um so I was thinking uh we should we should ship the the new pricing page on Thursday you know and then um send the announcement email right after.
It typed
I was thinking we should ship the new pricing page on Thursday and then send the announcement email right after.
Filler gone. The repeated false start collapsed.
You said
Hey team just a quick note that the deploy is done um everything looks green comma let me know if you see anything weird.
It typed
Hey team, just a quick note that the deploy is done. Everything looks green. Let me know if you see anything weird.
You said “comma” and got a comma, not the word.
You said
So um I need to add a a function called parse user config that takes a path and returns the settings struct and uh it should return an error if the file is missing.
It typed
I need to add a function called parseUserConfig that takes a path and returns the settings struct, and it should return an error if the file is missing.
The name is spelled the way you'd write it. It didn't write the function for you.
Selected in Mail
I need you to send me the report by tomorrow morning, no excuses. This is the third time I am asking.
You said
Make this shorter and more polite.
It typed
Could you please send me the report by tomorrow morning? I would appreciate it, as I am still waiting on this.
Text was selected, so what you said became an instruction about it.
Works in every app
No integrations, no plugins, nothing to keep up to date. Text arrives the same way it would if you pasted it. Only the style changes.
Slack
Gmail
Notion
VS Code
Cursor
Zed
Linear
Obsidian
Word
Discord
Terminal
Figma
…and anything else with a text cursor. These twelve are examples, not the list.
What it does
It knows where you're typing
The same sentence lands as a quick message in Slack, a real paragraph in Mail, and plain prose in an editor. Don't like a call it made? Pin the style for that app.
Rewrite by voice
Select text, hold the key, say what to change. Shorter, warmer, bullets, another language. Your selection gets replaced.
It learns your words
Add the names and jargon you use and they're spelled right. “Takumi” stops coming back as “tacoumi”.
Shortcuts for what you retype
Say a saved phrase and the full text drops in. Your address, a standard reply, boilerplate. A straight swap, not a guess.
Any language you speak
It works out what you're speaking. You can talk in one language and have it typed in another.
Nothing to set up
No API keys. No model picker. No provider account. No sign-up. Install it, pick a key, talk.
What we left out
A shorter list is the feature. We decided against these. We didn't run out of time.
No live transcript
Words crawling across your screen while you're still talking is something to watch, not something to use. You get the finished text, once.
Nothing running when you're idle
Between dictations there's no bar, no timer, and nothing of ours drawing on your display. Your laptop shouldn't know it's installed.
It always gives you something
If cleanup fails or takes too long, you get the plain transcript instead of an error. A rough draft beats nothing.
What leaves your machine
Your recording goes to a server to become text. We'd rather tell you exactly what that means than say “private” and stop there.
| Your recording | Uploaded, turned into text, gone. Never written to disk. Kept only if you turn that on. |
|---|---|
| Your words | Held in memory while they're cleaned up. Never stored on our side. |
| What we keep | Your word count, so the free limit and your stats work. Never the words. |
| Your clipboard | Read and put back on your machine. It never leaves it. |
| Analytics | None. Not in the app, not on this site. No trackers, no pixels, no third-party scripts. |
Free until it's worth paying for
The free plan is a real plan, not a countdown. When dictation becomes how you write, the paid one takes the ceiling off.
Pro
Unlimited$9.99 a month, or $99.99 a year
Unlimited dictation on three machines. Recordings up to 15 minutes. Vocabulary synced between them.
See what's included14 days to change your mind, any reason.
Questions first
Does it work offline?
No. Your recording goes to our server and comes back as finished text. Running the whole thing on your laptop would mean a much bigger download, a slower first run and worse output, so we took the other trade. What that costs you is written out above, in full.
Do I need an API key?
No. Nothing to sign up for and nothing to paste in. Install it, pick a key, talk. That's the setup, and it's most of what you're paying for.
What counts as a word?
Words that actually get typed. If cleanup drops an “um”, you don't pay for the “um”. The count resets on Monday and the app shows what's left.
Does it work in every app?
Anything with a text cursor. There are no per-app plugins — text goes in the same way a paste would. Only the writing style changes.
Do you keep my recordings?
No. Audio becomes text and is discarded. It's never written to disk unless you turn that on. We count your words so the free limit works, and never keep them.
What's different on Linux?
X11 only for now, and there's no tray icon. Same build and same behaviour otherwise.
Can I get a refund?
Within 14 days, any reason, just ask. The free plan needs no card anyway, so you can find out whether it fits first.
2,000 words a week, free.
No account, no card, nothing to paste. Install it, pick a key, talk. If it sticks, it's $9.99 a month.
macOS 12+ · Windows 10+ · Linux (X11)