A new kind of dictation

You speak – Pysar writes.

PIH-sar: stress on the first syllable, and "sar" as in commissar. Not "pie-sar".

Pysar turns speech into clean, punctuated text without sending your audio anywhere. Recognition runs on your own machine, so there is no account to make.

Download for macOS Download for Windows

The release is still being finished. Downloads will open here as soon as it's ready.

private · local · no subscription

Pysar

Pysar's main window

100% localaudio never leaves your device

Real timetext appears as you speak

Up to 100 languagesmixed speech, uninterrupted

macOS + Windowsthree editions for your hardware

Speak the way you actually speak.

Pick the languages you use and put them in the order you use them. Pysar follows that order instead of guessing at every sentence.

A term in another language dropped mid-sentence stays the word you said, not the closest thing that sounded like it.

And all of it on your own computer.

no account · no per-minute billing · no sending your voice to someone else's cloud

How it works

Three steps to finished text

01
Install and pick an edition

Download Pysar for macOS or Windows and pick the model that fits your computer. Everything runs offline.
02
Press the hotkey and speak

A compact widget appears. Speak in the language you think in, then switch mid-sentence when the right word lives somewhere else.
03
Text appears right where your cursor is

The finished, corrected text is inserted into your app. Stop the recording – and get back to work.

Living well takes strategy and tactics

To hold on to them you need a journal: every thought, every plan, every half-formed idea. And a journal like that is, in effect, classified.

So we built Pysar. It keeps that journal where it belongs, on your own machine. And it writes fast.

Fast and accurate

Speak for a minute –

≈3 sec

and the text is already waiting.

All figures are measured on the medium model, which is heavier and more accurate than the lightweight ones.

MLX · Apple Silicon (M4 Pro) measured · a minute in ~3 sec

24×

Nemotron 0.6B · regular CPU measured · a minute in ~3 sec

22×

WhisperX + NVIDIA CUDA estimated* · a minute in ~4 sec

~15×

WhisperX · Windows CPU estimated* · a minute in ~20 sec

~3×

Pysar listens while you speak, and the text lands instantly the moment you stop. Anywhere you can type.

Measured, not promised: Apple M4 Pro, July 2026. An asterisk marks an estimate for other hardware.

Everything worth keeping, kept

One press turns a dictation into a note. Notes sit on your own device; turn sync on and they travel to your other devices already encrypted - the key never leaves you.

Save any dictation as a note in one move.
Color-tag notes to find your way around fast.
Export to TXT, Markdown, or JSON whenever you need to take it elsewhere.
Optional note sync between your devices - through your own cloud folder or the Pysar server.

Notes

Notes screenshot – coming soon

Features

Everything dictation actually needs

Privacy isn't a setting here, it's the premise. Everything else is comfort: text lands wherever your cursor already is, and a change of language mid-sentence doesn't faze it.

Private by default

Recognition happens on your own machine. Not one byte of audio reaches a server, ours or anyone else's.

Live real-time dictation

The Nemotron edition puts words on screen while you are still talking. Nothing to wait for; the text arrives with the thought.

Up to 100 languages

Personal dictionaries and automatic corrections keep specialised terms and mixed-language speech clean from first word to last.

Paste into any app

The text lands wherever your cursor is - messenger, editor, browser, terminal. On Windows there is no automatic paste yet: the text waits in the clipboard.

No monthly rent

Bought once and yours to keep. You almost certainly have enough monthly bills without us.

History and notes

Every session is saved locally. Go back to earlier dictations, copy and edit them whenever it suits you.

Three editions for your hardware

WhisperX for maximum accuracy, MLX for Apple Silicon, Nemotron for instant streaming. Pick yours.

Auto-correction

Speech is never tidy

Real speech is full of false starts, corrections and thinking out loud. Pysar clears that up itself instead of leaving it for you.

"meeting on tuesday no on wednesday at three"

Meeting on Wednesday at three.

"call him back tomorrow morning period new paragraph

Call him back tomorrow morning.

"buy list milk bread eggs"

Buy:
– milk
– bread
– eggs

Instant

On by default, and it adds no waiting. It drops spoken corrections like "no, on Wednesday", restores punctuation, tidies numbers and times, and expands your snippets.

Full^*

Optional. It follows written instructions of your own: how to format, what to avoid, which tone to keep. It understands spoken editing commands – "new paragraph", "list", "delete the last sentence" – and commands you define yourself: say "signature" and your signature appears.

^* Full correction adds a few seconds of processing once you stop speaking.

Why Pysar?

The general scribe of the Zaporozhian Host was chancellor of the Cossack state and held one of the highest offices of the Hetmanate: he kept the state seal, ran the General Military Chancellery, and turned the spoken word into a document that carried weight. He was trusted with the state's correspondence and its secrets – and he knew how to keep them. We've put that office back to work. Now the scribe is on duty again: a golden quill at the ready, every word of yours lands on the line – no inkwell, no queue at the chancellery. And the office's core skill hasn't gone anywhere: what's entrusted stays kept. Everything you dictate stays on your computer.

Ilya Repin's painting Reply of the Zaporozhian Cossacks to Sultan Mehmed IV — Ilya Repin. Reply of the Zaporozhian Cossacks to Sultan Mehmed IV (1880–1891). The scribe gets the last word.

Editions

59 models. No need to choose.

Pysar recommends the three most accurate models for your chosen language and hardware – the ones that fit in your computer's memory.

There is a model for any machine, from one that barely touches your memory to the most accurate one. After a dictation Pysar gives the memory back: right away, after a few minutes, or never, whichever suits you. And when the system runs short, it gives it back without waiting.

Edition	Strength	Best for	Platform
WhisperX	Maximum accuracy	Long, high-stakes texts	macOS · Windows
MLX	Optimized for Apple Silicon	Speed without straining your Mac	macOS (Apple Silicon)
Nemotron	Live real-time streaming (experimental)	Dictation where instant matters	macOS · Windows

Three minutes from install to your first text

Pysar walks you through the setup and downloads the model itself. After that it works with no internet at all.

Settings

Settings panel

Start dictating today

Private and local. 30 days to try it, then $15 – a reduced price for early buyers.

Download for macOS Download for Windows

The release is still being finished. Downloads will open here as soon as it's ready.

Questions and answers

The essentials, briefly

Do I need internet to use it?

No. Once the model is downloaded, Pysar runs fully offline. Internet is only needed for the initial install and for updates.

Which languages are supported?

Pysar supports up to 100 languages across its available models. Choose the model that fits your hardware and language needs, then switch languages mid-sentence without losing your flow.

How much does it cost?

Pysar gives you 30 days with every feature, no card or sign-up required. After that it is $15 once – a reduced price for early buyers while Pysar is in beta. No subscription, no monthly charges.

How do I remove it completely?

Settings include a "Delete all data" action that clears downloaded models, history, and cache. If you used sync through our server, a second action next to it wipes your files from there too. Then remove the app itself the standard way for your system - nothing is left behind on disk.

What are the system requirements?

macOS (Apple Silicon or Intel) or a modern version of Windows; the MLX edition requires Apple Silicon. 8 GB of memory is enough for the fast Nemotron edition; for the larger WhisperX models, we recommend 16 GB.

Is it safe for sensitive data?

Yes. Audio is processed locally and never sent anywhere, which makes Pysar suitable for working with sensitive information.