Last updated: July 4, 2026
Privacy
LocalWispr collects nothing, because there is nowhere for it to go. There is no server, no account, and no analytics. Your voice never leaves your Mac.
Most dictation tools ask you to trust a privacy policy. LocalWispr was built so you don't have to. The speech recognition (the NVIDIA Parakeet model) and the optional AI polish (Apple Intelligence) both run on your Mac's Neural Engine. Your audio is turned into text on your own machine and then discarded from memory. This isn't a promise we make on top of a cloud service — it's the architecture. There is no cloud service.
What the app handles, and where it stays
Everything LocalWispr touches lives on your Mac. Nothing about it is uploaded, backed up to us, or shared.
- Your voice. Recorded only while you hold the hotkey, transcribed on-device, and dropped from memory once the text is inserted. It is never written to a server and never sent anywhere.
- Your transcripts. If history is on, your dictation history is stored as plain local files in
~/Library/Application Support/LocalWispr. You can turn history off entirely, or clear it, at any time. - Your personal dictionary and snippets. The words, names, and expansions you teach the app are plain local files in the same folder. They're used to improve your transcription on your Mac — nothing is learned by watching you and nothing is sent out.
- Your screen. LocalWispr never reads the text on your screen or in your windows. Cloud tools often upload window content as "context." LocalWispr does not.
The only two network requests the app ever makes
LocalWispr is functional in Airplane Mode. It reaches the network exactly twice, for two narrow purposes, and neither one carries your data.
The speech-model download
On first launch, LocalWispr downloads the ~1 GB speech recognition model from Hugging Face. This is a plain file download — the same file everyone gets. No user data is sent in either direction beyond the standard request needed to fetch a file. Once it's on your Mac, it's never downloaded again, and dictation works offline forever after.
The update check
LocalWispr asks updates.localwispr.com whether a newer version of the app exists. This is a version query and nothing more — it does not include your audio, your transcripts, your dictionary, an account, or any identifier about you. It's the same request a browser makes to load a page. You can turn automatic update checks off in Settings if you'd rather not make it at all.
That's the complete list. No audio, no transcripts, no screen content, and no usage data is ever uploaded — because the code to do so does not exist in the app.
No analytics. No accounts. No third parties.
- No telemetry or tracking code. There is no analytics SDK, no crash-reporting service, no ad network, and no third-party measurement code compiled into LocalWispr.
- No accounts. You never sign up, log in, or give us an email to use the app. Your $19 license is a key, not an account.
- No profiles. We don't build a profile of you, because we don't receive anything to build one from.
The website
This site, localwispr.com, is a static site served from Cloudflare. It sets no tracking cookies, runs no analytics or advertising scripts, and doesn't try to identify you. Standard server-level request logs handled by Cloudflare (the kind every website receives) are all that exists, and we don't use them to track individuals.
Questions
If anything here is unclear, or you want to know more about how a specific part works, email hello@localwispr.com and a real person will answer. LocalWispr is made by an independent developer operating as MEGAROOSTER LLC.