Privacy
Privacy Policy
Last updated: June 2026
Parleq is an open-source macOS dictation app made by Keavi LLC, a Virginia limited liability company ("Keavi," "we"). The short version: we don't collect your data.
We run no server and collect no dictation data. Keavi operates no backend that receives your audio, transcripts, or usage. Parleq runs entirely on your Mac.
Your audio never leaves your device. By default, speech recognition runs on-device on the Apple Neural Engine. Audio lives in memory only for the duration of a dictation — never written to disk, and never sent to Parleq or Keavi.
Transcript cleanup is on-device or your own provider. The optional cleanup step either runs on-device (processed entirely on your Mac, with no network call), or is sent to a cloud provider you configure (Google Gemini, OpenAI, AWS Bedrock, Google Vertex AI, or Azure OpenAI) directly from your Mac using your own account and credentials. Keavi is not a party to that exchange and never receives or stores that text; the provider's handling is governed by your agreement with them. You can also skip cleanup entirely.
Voice enrollment stays on your device. The optional voice-enrollment feature — which teaches Parleq to tell a term apart from a similar-sounding word by the sound of your voice — runs entirely on-device. What's kept is a mathematical voiceprint (an embedding) — not a transcript. Voiceprints are biometric data: they're stored encrypted (AES-GCM) under a key in your Mac's Keychain that is device-only and never synced to iCloud, are enrolled only for the terms you opt in to, and can be deleted per term or all at once. By default, Parleq also keeps those short enrollment clips — encrypted (AES-GCM) on your Mac under that same device-only key — only so your voiceprint can be re-derived after an on-device model update; you can turn clip storage off at any time, which erases them. Nothing about enrollment or the live disambiguation is sent to Parleq, Keavi, or any cloud.
No telemetry in the app. Parleq contains no analytics, crash-reporting, advertising, or tracking SDKs and does not phone home. Its only outbound connections are ones you initiate: update checks, your chosen cloud cleanup provider, and one-time model downloads from Hugging Face (the on-device speech model, and the on-device cleanup model if you enable it).
This website. parleq.app uses GoatCounter — a privacy-friendly, open-source analytics tool that sets no cookies, needs no consent banner, and records only aggregate pageview and referrer counts. No personal information, no cross-site tracking.
Downloads. The app is distributed via GitHub; on-device models are downloaded from Hugging Face at your request. These third parties have their own privacy practices.
Changes. We'll revise this page and its date if our practices change.
Contact. Questions? Email hello@parleq.app.
Keavi LLC · A Virginia limited liability company · keavi.app