OpenASR Desktop is now on Windows

OpenASR Desktop runs on Windows today. If you have been using it on macOS, everything you know transfers — same models, same local HTTP API, same privacy boundary. If Windows is where you are starting, this is what to expect.

What you get

The same feature set that ships on macOS:

• File transcription — drop in audio, get a transcript with SRT/VTT subtitle export.
• Dictation — real-time text insertion into other apps.
• Live captions — speech transcribed as you speak, on your machine.
• Speaker diarization — anonymous who-said-what labels, session-scoped, embeddings discarded at the end.
• System audio capture — transcribe what your computer is playing, meetings included.
• Hot words — bias recognition toward names and terms that matter to you.

Twenty-six models are available through the signed catalog. Everything runs locally. No audio leaves your machine.

Requirements

• Windows 10 or later, x64.
• The app is code-signed via Azure Trusted Signing (publisher: Caelum Inc.).

The open-source Rust core underneath is the same Apache-2.0 codebase that runs on macOS and Linux. The desktop shell is closed-source, but adds no hidden transcription path — everything that touches your audio is the open code you can read at github.com/QuintinShaw/openasr.

About the SmartScreen prompt

Windows SmartScreen may show an “unrecognized app” warning the first time you launch OpenASR. This is normal for a new publisher identity — SmartScreen builds reputation from download volume, and a day-one app has none yet.

To proceed:

1. Click More info in the SmartScreen dialog.
2. Click Run anyway.

That is it. The app is signed with a valid certificate from a trusted certificate authority; SmartScreen is not questioning the signature, just the newness of the publisher. As more people download and run it, the warning disappears on its own.

We would rather explain this upfront than have you wonder whether something is wrong. Nothing is.

Downloads and mainland China acceleration

Download the Windows installer from the official download page. Users in mainland China are automatically routed through an accelerated channel — no configuration needed.

Where to go from here

Install, open, pick a model, and talk. If something breaks or feels off, open an issue on GitHub — that is where we track everything, and it is the fastest way to get a response.

If you find it useful, a star on GitHub helps more than you might think. It is how other people find the project.