Hold a key.
Speak your mind.
Dictation and meetings, on your own machine. Hold a key to talk into any app, or record a call and get both sides transcribed as they speak. No cloud, no account, nothing to pay.
Nothing typed yet.
Tap to run it. Or hold it, the way you would hold the real key — Right Ctrl and Space on your own keyboard work too.
You think four times faster
than you can type.
Speech runs about 150 words a minute. Typing runs about 40. Dictation takes your hands out of the path between the thought and the text, so the sentence lands while you still have it.
general figures for speech and keyboard rates, not a benchmark of this app
However you use it,
nothing leaves your machine.
Hold the key, say what you mean, let go. Clean text lands where you were typing.
Hold a key. Talk.
One system-wide hotkey, Right Ctrl by default. Hold it, talk, release. You stay in the app you were already typing in.
one key watched · every other keystroke ignored
One thing: whether your push-to-talk key is down.
It lets every other keystroke through and stores nothing you type.
Text appears while you speak.
Words appear while you are still talking, not once you stop. You can watch it keep up.
runs on your own processor · no graphics card needed
Each update extends the last instead of starting over, so the words keep pace with you instead of arriving in a lump at the end.
It lands cleaned up.
Fillers, stutters and mid-sentence corrections come out. Punctuation and capitalisation go in. Then Swar types the result into whatever had focus, whether that is your editor or a Slack box.
cleaned up on your machine · no plugins, no integrations
I told the team um the the new timeline should be ready by Friday sorry Thursday
I told the team the new timeline should be ready by Thursday.
One key, held. Or the whole call.
- you0:07
I have got the recorder on, and it is picking up both sides.
- them0:19
Can we start with the encoder question?
- you0:44
The hardware path works on both machines.
- them1:12
What does it cost if we turn it on?
Nothing.
Not a byte.
No audio, no transcripts, no usage or crash reports. There is no switch to leave off, because Swar has nowhere to send any of it.
Watch it yourself with any network monitor, or run the app behind a firewall with everything blocked. You get the same list.
Watch it yourself with any network monitor, or run the app behind a firewall with everything blocked. You get the same list.
Bring your own model.
We are not the gatekeeper.
Forever, and there is no second card.
Free to install and free to use, running on hardware you own against a model on your own disk. No servers run it, so no bill reaches you.
- no account to create
- no seats to buy
- no usage meter
- no free trial that ends
- no per-minute transcription bill
- no minutes to buy before a meeting
- no team plan gating the good parts
- no telemetry paying for it quietly
- no card on file
The questions
a sceptic asks.
Including the ones with an awkward answer. Anything that installs a keyboard hook and types into your applications should expect them.
Does any audio leave my machine?
No. There is nowhere to send it. Swar reaches the network once, on first run, to download the speech model, and never again. Outbound access is switched off at the boundary, so it stays that way even if a future build adds a call. You cannot leave this off by accident, because there is no switch.
So it is offline?
Offline after setup. It reaches the network once during setup, to download the speech model, which Swar checks against a known fingerprint before installing. Copy that folder across from another machine and it never reaches out at all. Whether the model ships inside the installer is still open.
It installs a keyboard hook. Should that worry me?
You should want an explanation before installing anything that does. Swar asks one question: is your push-to-talk key down. Every other keystroke passes straight through untouched, and nothing you type is stored, buffered or sent anywhere.
In a meeting, how does it know who said what?
Them and you, and it is not a guess. Your microphone is one recording; this computer’s own output is the other. Swar never mixes them before transcribing, so the side a line came from is a fact about how Swar captured it, not a model’s opinion about a voice. One limit comes with that: everyone on the far end shares one channel, so a four-way call gives you your side and everybody else’s in one bucket.
Does it tell people they are being recorded?
No, and it will not do it on your behalf. Swar never starts on its own: no calendar trigger, no watching for meeting apps. A recording exists because you started one. Before the first one, Swar asks you to confirm that telling everyone is yours to do. While it records, the tray icon stays red, and no setting turns that off.
Does it summarise the meeting?
No. Swar records, transcribes both sides during the call, and lets you search every meeting you have had. A local pass could pull out common words and phrases shaped like commitments, but that is not what people mean by a summary, so Swar does not show one. It does keep a record of anything it failed to capture.
Where do transcripts and recordings live?
On your own machine. Dictation history goes in a database; each meeting gets a folder holding its two audio tracks and the screen recording, if there was one. They expire on separate clocks you control: recordings after a week by default, text after three months. One button deletes the lot. Swar uploads nothing and trains nothing.
Can I use my own model?
Yes. Swar is not tied to one company’s model. Whisper and the other open speech models work, in whichever language they cover. Point Swar at a model folder and it runs.
What does it cost?
Nothing, and there is nothing to upgrade to. Swar is free to install and free to use, running on hardware you already own, with no account, no card and no usage meter.
Which platforms?
Windows 11 today, macOS next, both halves of it. Recording what this computer is playing uses a Windows interface with no drop-in equivalent elsewhere. Download it, run the installer, pick your hotkey, and you are dictating.
Stop typing what
you could say out loud.
Free forever, and it never phones home. Windows 11 today, macOS next.
no account · no card · no telemetry