A security overview, for the people who have to sign it off.
Mythral is dictation and meeting notes for Mac. Speech recognition, writing and notes run on Apple silicon with local models. There are no Mythral servers, no accounts and no telemetry, so there is no customer data for us to hold, lose or hand over.
Architecture
- Speech to text runs on the Mac: NVIDIA's Parakeet model on the Neural Engine for English, and whisper.cpp for 98 other languages.
- Writing and meeting notes come from an open model (Qwen3.5) running on the Mac through llama.cpp.
- Meetings are heard through macOS's own audio capture. No bot joins the call.
- The app is signed with a Developer ID and notarized by Apple.
Every connection the app can make
An organization can turn all of these off with one setting, OfflineMode.
What is stored, and where
Deleting a meeting deletes its transcript, notes and audio. Removing the app and its folder removes everything; there is no copy anywhere else.
Controls for organizations
Every setting can be set and locked with a configuration profile from your MDM, including offline mode, updates, meeting recording, a consent notice before each recording, and how long audio is kept. See Mythral for teams.
Permissions
- Microphone, to hear the person dictating.
- Accessibility, to type into the app in front, and to read a little of the screen as spelling hints for that one dictation. Password managers and banking apps are never read.
- Screen & System Audio Recording, only for meetings: it is how macOS lets an app hear the other side of a call. No screen is recorded.
Requirements
Apple silicon Mac · macOS 14.4 or later. 16 GB of memory recommended for meeting notes.
Questions
For a security questionnaire, a call with your security team, or anything not covered here, talk to us. Full detail on the app's data is in the privacy policy.