Whisperstream

A closer look at Whisperstream

Whisperstream is private, on-device push-to-talk dictation for Windows and Mac: speak anywhere you can type, and transcription runs entirely on your computer, so your voice never leaves your machine. It is free and open source, built on NVIDIA Parakeet, with AI cleanup that cleans up what you said, a Stream-1 Mini model that corrects when you change your mind, and Agent Speak for real-time voice for coding agents, while core transcription uses open-weight speech models including Parakeet, Qwen3 ASR, or Cohere Transcribe depending on language and accuracy needs.

Editorial guidance based on published documentation. The workflow below is a suggested starting point, not a measured product result.

When to choose Whisperstream

• People who want dictation without sending voice to the cloud, since Whisperstream transcribes entirely on a Windows PC or Apple Silicon Mac and says your audio is never sent to the cloud for core transcription.
• Fast thinkers who type slower than they speak, since Whisperstream's pitch is that typing caps you at 40 to 60 words a minute while you think and speak roughly three times faster.

When to choose something else

• Users on Linux or ChromeOS, since Whisperstream is presented as dictation for Windows and Mac, with core transcription on Windows PC or Apple Silicon Mac.
• Users who want a fully cloud-free product with no internet use at all, since the features page says model downloads, official-installer updates, and optional cloud, Agent Speak, or external-destination features use the internet as described in its documentation.

A practical starting workflow

1. Download Whisperstream for Windows or Mac from the homepage, or star it on GitHub.
2. Press to talk and speak anywhere you can type, letting on-device transcription turn speech into text.
3. Let AI cleanup clean up what you said, and let Stream-1 Mini correct the text when you change your mind.
4. Use Agent Speak for real-time voice for coding agents, and keep core transcription local so your voice never leaves your machine.

Check before committing

• The homepage says download free and free and open source, with a yearly-cost comparison of $0 against $140 or more per year for a cloud path, but no plan, subscription, or billing terms appear in the text read, so no free allowance fact is recorded in this pass.
• The directory tagline mentions a one-time $29 fee, but no $29 amount appears in the current homepage or features text read, so that older figure is not repeated as a current price.
• Core transcription runs on your computer, but model downloads, installer updates, and optional cloud features use the internet, according to the features page.

Common questions

Compare approaches for this task

• Willow Voice — People who type for work and chat all day, since Willow works anywhere you type, on Mac, Windows, and iPhone, with dictation plus Scribe writing from intent.; Teams with strict privacy needs, since the Business tier adds enforced privacy mode with zero data retention and SOC 2 Type compliance.
• Wispr Flow — People who write in many apps and hate typing, since Wispr Flow turns speech into clear, polished writing in every app, on Mac, Windows, iPhone, and Android.; Teams that need notetaking plus admin control, since paid plans add unlimited Notetaker access, team management, admin controls, and support.