Wren Get WREN
Dictation for Windows

Hold a key. Talk.
Your words land where
your cursor is.

Hold ctrl+win

Everything happens on your own machine — the listening, the transcribing, your words. Nothing is ever sent anywhere.

$49 onceno subscriptionno accountworks with the network off

The app

Open it and it tells you what a week of talking was worth.

38,209Words dictated 3,883 today
9.6hTime saved vs typing at 45 wpm
140wpmYour speaking speed 3.1x your typing speed
375msTypical wait 9 in 10 under 798ms
Words each day
5,260Sat 5,861Sun 3,842Mon 3,028Tue 3,497Wed 548Thu 5,146Fri 5,530Sat 3,883Sun
Speaking
Dictations1,817
Words each time21 words
Time spent talking4.6 hours
Busiest day 2026-09-27 · 5,861 words

Time saved compares your measured speaking time against typing at 45 words a minute. Nothing here is estimated except that one number.

Your words
linked inbecomesLinkedIn
my sequelbecomesMySQL
doctor patelbecomesDr Patel
kuber netesbecomesKubernetes
post gresbecomesPostgreSQL
jay sonbecomesJSON

38 words WREN writes your way. The file reads cleanly.

CloseQuit GPU · 17.5s

Not a screenshot — that is the real interface, drawn here in the same code the app uses. The figures are a real week of building WREN with WREN.

$49 once. You own it. No subscription, no seat, and nothing to cancel. Version 1 gets fixes for ever — if a Windows update breaks it, repairing that is ours to do, not yours to buy a second time.
Offline beats private. No account, no upload, no server doing the work — and today no networking code in it at all. Pull the cable out and nothing changes, which is the only honest test there is.
Your words follow you. Keep your vocabulary in a folder Dropbox, OneDrive or Drive already watches, and it is waiting on your other machine. No service of ours in the middle, and nothing more to pay.
Two machines

Your words follow you.
Nothing else does.

Everything WREN knows about how you speak is one plain text file. Put it in a folder something already syncs, and it is on your other machine the next time you open it.

DropboxOneDriveGoogle Drive Proton DriveiCloud DriveSyncthing Nextclouda git repoa network share

No account of ours in the middle. No service to pay for. No integration to go wrong, because there isn’t one — it is a file in a folder.

Only the words travel. Your history, figures and settings stay on the machine that made them, deliberately. Two computers writing the same file seconds apart is how a cloud starts leaving copies behind.
It catches cloud conflicts. Edit on two machines at once and Dropbox leaves a second copy beside the first, silently. Your work is not lost, it is somewhere nobody looks. WREN names those files on screen.
Unplug it and it still works. If the folder is not there, because a drive is out or a client signed itself out, WREN falls back to its own copy and carries on. It never goes wordless and never goes quiet about it.
Your words

It learns the words that matter to you.

Every tool gets names and jargon wrong. WREN keeps a plain text file of yours and applies it every time. You can open it in Notepad.

You say
WREN writes
linked in
→
LinkedIn
my sequel
→
MySQL
doctor patel
→
Dr Patel

It is a table, not a guess. The same words come out the same way every time, and nothing in that step can invent anything.

The engine

WREN runs offline.
The engine is GLOSS.

No model in the path to invent anything, nothing to upload, and a vocabulary you can open in Notepad.

GLOSS The local vocabulary engine inside WREN.

Built by Vonfal AI.

GLOSS is the part that turns what you said into what you meant — your vocabulary, your capitalisation, the filler taken out. It is a table you can read, applied in well under a millisecond, with no model in the path to invent anything.

And because the whole of it runs here, there is nowhere for your words to go. No account. No upload. No server doing the work. Today WREN contains no networking code at all — nothing in it could open a connection if it tried. It keeps the text of your last fifty dictations on your own disk so one that landed in the wrong window is not lost, and your audio is never written to a file at all.

Speed

Fast enough to stop thinking about.

Measured in ordinary use, not on a benchmark.

375mstypical wait after you let go
798msnine dictations in ten, under this
0.063msto apply every word you have taught it
0.19sa seven-second phrase, no graphics card at all

The first two are the most recent 500 of 1,960 real dictations on a desktop PC — the same figures WREN shows you on its own Stats screen, worked out the same way. Your machine will differ. A graphics card helps and is not required — the last number was measured with one switched off. We do not publish an accuracy percentage, because we have no honest way to measure one.

Get WREN

Seven days free. Buying is closed right now.

Nothing is wrong with WREN and nothing is wrong with your copy. The shop is switched off for a moment and will be back. If you already own it, your key is on your own machine and is not affected by any of this.

WREN 1.0 Seven days free, no card, no account
Windows10 and 11 · 64-bitAvailable now
macOSComing soon
LinuxComing soon
Download WREN for Windows

Version 1.0 · 64-bit · Windows 10 and 11 · the installer is not published yet

A key will be $49, once, when buying reopens. Version 1 gets fixes for ever.

Version 1 gets fixes foreverIf it breaks on a Windows update, that is ours to fix. A repair is not a new version.
Version 2 will cost againNew features live in the next version. We would rather say that now than call a paid upgrade a subscription later.
90-day graceBuy within 90 days of a new major version and you get it free, so nobody is punished for buying at the wrong moment.
Questions

The things people ask first.

Does it really work with no internet?
Yes, all of it — and more completely than that question usually means. The listening, the transcribing and your vocabulary all happen on your own machine, and today WREN contains no networking code whatsoever: there is nothing in it that could open a connection even if it wanted to. A later version will add one optional check for a new release. It will be a switch you can see, and this page will say so before that version ships.
Is my voice recorded or kept anywhere?
No. The audio exists in memory while you hold the key and is gone when the words are written — it is never saved to a file, on your machine or anywhere else. What WREN does keep is the text of your last 50 dictations, on your own disk, so one that landed in the wrong window is not lost. You can clear that whenever you like.
Do I need a graphics card?
No. A seven-second phrase takes about 0.19 seconds on an ordinary processor with no graphics card involved at all. One helps, and it is not required.
Which languages does it understand?
English, at the moment. That is a real limit rather than a soft one — the model WREN ships with is the English one — so if you dictate in another language this is not yet the tool for you.
How does the cloud syncing work?
You point WREN at a folder your cloud already watches — Dropbox, OneDrive, Google Drive, anything that syncs a folder — and keep your words file there. It then follows you to your other machines. There is no service of ours in the middle and nothing extra to pay. WREN also tells you when your cloud has quietly made a conflicted copy, which most of them do without saying so.
What exactly am I buying for $49?
The app, for as long as you want it, on your own machine. Version 1 gets fixes forever: if a Windows update breaks it, repairing that is ours to do. New features will live in a version 2 that costs again — we would rather say that plainly now than call a paid upgrade something else later. Buy within 90 days of a new major version and you get it free.
Is there a Mac version?
Not yet. Windows first, deliberately — local dictation on the Mac is a crowded shelf and on Windows it is nearly empty.
Can I try it before buying?
Not yet. WREN is finished and in daily use, but it is not packaged for other people's machines, so there is nothing honest to hand you today. When there is, it will be on this page.