Doesn’t running several models make it slower?
The opposite. The models run in parallel, not one after another, so the round trip is only as long as the fastest useful answer. When one provider has a slow moment, ZWhispr proceeds without it. That is what shrinks the worst-case P95 and P99 latency you actually feel.
How does combining models reduce errors?
Different models make different mistakes. ZWhispr merges their transcripts intelligently, using your dictionary and the context on your screen to pick the right reading wherever they differ. A slip from one model gets corrected instead of ending up in your email.
What exactly does the screen context feature see?
When you enable it, ZWhispr takes a screenshot of the window you are working in at the moment you start speaking and extracts the distinctive words on it. Those words are sent along with your audio as vocabulary hints. It is optional and you grant the permission in macOS.
Do I have to choose or configure a model?
No. ZWhispr always uses the current state-of-the-art models from leading third-party providers and upgrades them for you. There is nothing to select or tune.
What do you store?
We never store your transcripts or your screen context. Both are used to produce the transcription in front of you and then discarded. We do store your dictionary words and the account and usage metadata needed to run the service.
How is billing handled?
Subscriptions are $19 per month, billed through Paddle, our merchant of record. Paddle handles payment, invoices, and sales tax for your country. You can update your card or cancel any time from the billing portal link in your receipt. Questions go to hello@zwhispr.com.
Which apps does it work in?
Any Mac app with a text field. Hold the hotkey, speak, release, and the text is typed at your cursor in Slack, Mail, Notion, Cursor, Xcode, the terminal, a browser, or anything else.