AI Policy

Recall · SpeedUI.ai LLC · last updated 17 October 2026

The short version. Every AI feature in Recall runs on your iPhone, using Apple's built-in on-device technology. Nothing you say or write is sent to us, to Apple's servers or to any AI company, and nothing is used to train AI models.

Where Recall uses AI

Transcription — Apple's on-device speech recognition turns your recordings into text, live while you record or afterwards. Recall requires on-device recognition and never falls back to Apple's speech servers.
Titles, summaries, tags and folders — Apple's on-device language model (Apple Intelligence, through Apple's Foundation Models framework) reads a note and suggests them. On iPhones without Apple Intelligence, simple keyword rules inside the app do it instead.
Telling speakers apart — after a recording, Recall works out who spoke when and labels the transcript "Speaker 1", "Speaker 2" and so on. This runs on your iPhone with two small speech models that ship inside the app; no audio is sent anywhere. It doesn't know who anyone is, only that the voices differ.
Note pictures and emojis — each note can get an illustration made by Apple's Image Playground, which runs on your iPhone with Apple Intelligence, and an emoji picked by the on-device model (or a built-in list). You can remove a picture, make a new one, or turn pictures off in Recall's settings.
Emojify — the same on-device model picks words in your note that suit an emoji. It only adds emojis next to words; it never changes your words. Without Apple Intelligence, a built-in word list is used.

What we don't do

No cloud AI, no third-party AI services, no AI accounts or API keys.
Your notes, recordings and transcripts are not uploaded, stored by us, shared or sold.
Your content is never used to train or improve any AI model, ours or anyone else's.
No AI-generated content is shown to anyone but you.

Accuracy and control

AI output can be wrong: a transcript may mishear a word, and a summary or folder may miss the point. Everything AI produces in Recall can be edited — rename a note, change its folder, edit the transcript, tap Re-transcribe or Re-organize, or Remove emojis. The original recording is always kept, so nothing is lost if a transcript is wrong.

Turning it off

You can deny Speech Recognition access in Settings → Privacy & Security; recordings are then kept without transcripts. Apple Intelligence can be turned off in Settings → Apple Intelligence & Siri, and Recall then uses its keyword rules.

Credits

Speaker recognition uses FluidAudio (Apache 2.0) with the LS-EEND model by Di Liang & Xiaofei Li (Westlake University), in a Core ML conversion by GradientDescent2718; licensed MIT.