AudioToText

iOS does not allow a keyboard extension to use the microphone directly. AudioToText works around that by having the main app claim the microphone and hand the transcribed text to the keyboard through a small piece of storage shared between them — and iOS only allows a keyboard to read that shared storage once Full Access is granted.

Full Access is used for that one purpose. It is not used for network access, analytics, or anything else. See the Privacy Policy for the full explanation.

Only once per session. The first time you dictate, AudioToText opens briefly to claim the microphone — that's an iOS requirement, not a choice. After that it stays running in the background, and the keyboard's mic button starts and stops listening in place, in whatever app you're typing in, with no further switching.

Yes, once a model is ready. Apple's built-in Speech engine needs no download at all. The optional NVIDIA Parakeet models are downloaded once, then work fully offline from then on.

Apple's built-in engine is the default: no download, and it supports every language your device does. The Parakeet models are optional, English-only alternatives with different accuracy/speed trade-offs — pick whichever fits how you use dictation, from the Model screen in the app.

No. Transcription and AI summarization both run entirely on your device. See the Privacy Policy for the full picture, including the one place the app does use the network — downloading an optional model's files, not your data.

They stay in the app's local storage on your iPhone. Deleting a recording removes it; deleting the app removes everything at once. There is no server copy anywhere.