Skip to content

Releases: vishaldesai/Swar-install

Release list

Swar 0.1.0

Swar 0.1.0 Latest
Latest

Choose a tag to compare

View all tags
@vishaldesai vishaldesai released this 28 Aug 00:37
v0.1.0
2c5bb13

Hold Right Ctrl, talk, release, and cleaned-up text is typed into whatever had focus.
The speech model runs on your machine.

Take Swar_0.1.0_x64-setup.exe unless someone has told you to take the .msi.
Verify the SHA-256 below before running it, and read the
install steps — the build is
unsigned, so Windows will stop you once.

The first public build. Dictation works end to end: hold Right Ctrl, talk, release,
and cleaned-up text is typed into whatever had focus, with live partial text on the
overlay while you are still speaking.

In this build

  • Push-to-talk dictation with a configurable key, streaming on-device speech recognition,
    and text injection into whatever application has focus.
  • Seven selectable speech models across three tiers. The default,
    nemotron-3.5-asr-streaming-0.6b, streams cache-aware, so text appears before you
    release the key.
  • The Hub: dictation history, a personal dictionary, snippets, settings, and a
    notification centre.
  • Voice Profile — the dictionary learns the words you actually use.
  • Insights — words per minute, streaks, per-application usage, and a scratchpad.
  • Local storage only. One SQLite file under %APPDATA%\Swar\, a retention setting you
    control, and one button that deletes every transcript.

Assets

File Bytes SHA-256
Swar_0.1.0_x64-setup.exe 10,605,906 60ac22a56b82c0d9e61431d08ab347c720a3548b3181ceac2b5bf1a252869108
Swar_0.1.0_x64_en-US.msi 14,913,536 f74b72185e6ac140c8517fa29a173f9f694d2fb3ec7fd71afce36667cbc331bd

Known limitations

  • Unsigned. SmartScreen warns on the .exe; the .msi shows an unknown publisher.
    Verify the checksum above before running either. A signing certificate is next.
  • No updater. Nothing checks for a new version. Watch this repository's releases.
  • Windows 11 x64 only. No macOS or Linux build.
  • The speech model is not bundled. First run downloads 453.3 MiB from the
    k2-fsa asr-models release, SHA-256 pinned and installed fail-closed. After that,
    Swar does not touch the network.
  • The MSI upgrade code is derived, not pinned. A future release may not upgrade a
    0.1.0 MSI install cleanly. Uninstall before installing a newer .msi if it refuses.

Not built yet. The cleanup tier, transforms, and the meeting notetaker are designed
but unbuilt — each is blocked on whether a local language model earns its place, not on
effort.

Assets 8
Loading