Privacy First

Privacy Policy

Last updated: 2026-09-07

PrivaSpeech is built on a foundational principle: your voice data belongs to you alone. All processing happens locally on your device. We do not upload your dictation audio or transcripts to our servers.

Quick Summary

On-device transcription. Your dictated audio and transcripts are processed locally.
No cloud transcription. We don't send your audio or transcripts to third-party transcription APIs.
Imported audio stays local. Files you choose to import are decoded and transcribed on-device. Supported direct imports are `.wav`, `.mp3`, `.m4a`, and `.aac`.
Setup requires downloads. Models are downloaded over HTTPS on first launch (or after updates) and then stored locally for offline use.
Website analytics & ads. This website uses PostHog for pageview and website interaction analytics (respects Do Not Track) and Meta Pixel to measure ad campaign effectiveness. Advertising tags run on the website; optional desktop setup analytics are described below.
Privacy-first clipboard. Updates happen locally; we never read your existing clipboard.
Secure payments. Processed via Polar/Stripe; we never see full card details.

Data We Handle

Audio: captured only while recording, or read from a local file you explicitly choose to import. Microphone audio is stored temporarily as a WAV file in your temp directory, then deleted after transcription.

Transcripts: handled locally. Microphone transcriptions are copied to your clipboard; imported-file transcriptions do not auto-copy by default. We do not transmit transcript contents.

History: if you keep transcription history enabled, entries are stored locally on your device. Imported files can be run in private mode to skip history for that import.

Device information: limited to what Windows and Tauri need to register hotkeys and access the microphone. Optional setup analytics use a random app identifier. Network services receive connection metadata when used.

Model downloads: fetched from Cloudflare R2 on first run or update, verified with checksums, and stored permanently on your device for offline use. No subsequent internet connection required for transcription once models are installed and the license is activated.

Payments & Licensing

Purchases are processed by Polar using Stripe. Payment details are entered on their hosted pages, not in the app or on our servers. We receive basic transaction metadata (like email for receipts/support) to deliver licenses; we do not receive full card numbers.

Crash Reporting & Analytics (Desktop App)

Setup analytics are off by default. In builds with this option, enable “Share anonymous setup milestones” during onboarding or in Settings. PostHog receives a random app identifier and bounded setup milestones: startup, model download, microphone check, license activation, and first dictation. Audio, transcripts, clipboard contents, file paths, emails, and license keys are not included.

Turn setup analytics off in Settings to stop future milestone delivery and remove the local analytics identifier. This does not delete events already received. Network services receive connection metadata; desktop analytics disables GeoIP enrichment. Older releases may not offer these controls. There is no automatic desktop crash-report upload.

Model downloads, license validation, and update checks use the network independently of optional analytics.

Website Analytics & Advertising

Our marketing website uses PostHog to understand aggregate traffic and website interactions (such as pageviews, clicks, checkout events, purchases, and other navigation events). We do not enable session recording. PostHog is configured to respect Do Not Track(DNT); if your browser sends a DNT signal, analytics collection is disabled.

We use this website analytics data to understand which pages, calls to action, campaigns, and content paths are most useful. Paid-search attribution can include search ad click IDs such as `gclid` or `msclkid` and UTM campaign parameters. We do not use PostHog session replay on the website.

We also use Meta Pixel to measure the effectiveness of our advertising campaigns on Meta platforms (Facebook, Instagram). This collects:

Checkout initiation events when you click pricing options
Standard browser/device information for attribution

Meta Pixel data is processed by Meta according to their Privacy Policy. You can opt out of Meta's interest-based advertising via your Facebook Ad Preferences or through industry opt-out tools like Digital Advertising Alliance.

Important: Advertising tags run on the marketing website. Optional desktop setup analytics are described above. Dictation audio and transcripts are not sent to analytics.

The website also loads the Google Ads base tag for advertising attribution. Checkout return pages are not treated as paid conversions. Verified Polar webhooks report trial/subscription creation separately from paid orders, using hashed customer and order identifiers. These identifiers are pseudonymous; server billing events are independent of browser Do Not Track.

Your Controls

Complete data ownership. All data stays on your device. Delete models anytime.
Private imports. Use private mode for imported files when you do not want that transcript written to local history.
Separate billing account. Polar uses your checkout email for receipts, licensing, and customer portal access.
Uninstall anytime. Remove the app completely; user data remains under your profile for reuse if you reinstall.
Microphone control. Disable microphone access in Windows privacy settings to stop audio capture.
Works offline. Turn off Wi-Fi and PrivaSpeech can continue transcribing after models are installed and the license is activated.

Contact

Questions or privacy requests? Email support@privaspeech.com.