How AI works in Magic Morning
One sentence: Magic Morning's AI runs on your iPhone using Apple Intelligence. No prompt, no data, no output ever reaches our servers — we don't have any.
1. Why this page exists
Art. 50 of the EU AI Act (Regulation (EU) 2024/1689) requires that users of AI systems are informed clearly that they are interacting with an AI system, and what it does. This page is that disclosure for Magic Morning. It is intentionally readable; the legal references stay short.
2. What the AI in Magic Morning does
Magic Morning's AI features generate four categories of text, all surfaced inside the app with a ✦ AI badge that cannot be removed:
- Weekly Recap — a short narrative summary of your past week, generated Sunday evening (or when you open it manually).
- AI Insights — pattern observations across your habits and routines, generated when you have at least two weeks of activity.
- Identity Framing phrases — the "you are someone who …" line attached to a habit.
- Habit Triggers — suggestions for when, where, or after-what to perform a habit.
3. Where the AI runs
Inference runs on your iPhone in Apple's Foundation Models framework, which is part of Apple Intelligence. The model is provided by Apple and runs in the device's Neural Engine.
No prompt, no envelope, no insight, and no recap is sent to a server. Magic Morning has no backend, no proxy, no third-party AI API. If you turn off Wi-Fi and cellular, the AI features still work.
4. What the AI is allowed to read
Only aggregated activity from this device, subject to the per-category toggles in Settings → AI features:
- Counts and timestamps of routine runs and habit logs
- HealthKit aggregates — average sleep hours, total workout minutes per day, average step count, mindful minutes
- Calendar event timing and busy hours
- Your display name for personalisation
You can revoke any of these categories individually. Revoking "Health" stops the AI from reading any HealthKit aggregate even though HealthKit access itself stays granted; same for the other three categories.
5. What the AI never reads
- Raw HealthKit samples (only aggregates)
- Calendar event titles, attendees, locations, conferencing links, notes
- Contacts, photos, messages, mail, browser history
- Anything from outside Magic Morning
6. When AI isn't available
Apple Intelligence requires compatible hardware (an iPhone with Apple Intelligence enabled). On older devices, on devices where Apple Intelligence is turned off in iOS Settings, or on devices where the model has not finished downloading, AI features are silently disabled. You still get insights — produced by deterministic rules over your habits and routines on-device, no model required.
7. How to disable
Settings → AI features → AI features → Off. The toggle hides AI surfaces immediately. Per-category toggles (Routines, Habits, Health, Calendar) let you revoke individual data categories so the AI can't read them, even on-device.
8. How to recognise AI content
Every AI-generated insight, recap, or suggestion shows a ✦ AI badge. Editable AI-suggested fields (Identity Framing, Trigger) carry an "✦ AI suggestion" label until you confirm or override them. The badge is non-removable and present even if you change theme or font size.
9. Limitations and risks of AI output
The on-device model is a language model. It is good at fluent summarisation and pattern-naming. It is not:
- a medical-grade observer — its insights are not therapy, not diagnosis, not medical advice.
- peer-reviewed against published studies — copy may reference common habit-formation heuristics, not citations.
- infallible — small or unusual data sets can lead to pattern claims that don't really exist. We mitigate by suppressing any insight whose numbers don't trace back to your actual data, but we cannot promise zero hallucinations.
Treat AI output as orientation, not as instruction. The Terms of Service, Section 5, formalises this.
10. AI risk classification under the EU AI Act
Magic Morning's on-device AI is a general-purpose AI assistant for personal productivity. It is not on the EU AI Act Annex III high-risk list, and it does not perform biometric identification, emotion recognition, credit scoring, or any other high-risk function. It is subject to the Art. 50 transparency obligation (which this page satisfies), not to the high-risk obligations in Chapter III Section 2.
11. Provider responsibility
The Apple Foundation Models model is provided by Apple Inc. as the upstream provider. Magic Morning is the deployer (Betreiber) under Art. 26 EU AI Act. We take responsibility for the way the model is prompted and how its output is surfaced inside the app — including the data minimisation choices in Section 4 above and the non-removable AI badge.
12. Changes to this disclosure
If we add new AI features or change how data flows into the model, we will update this page and the in-app explanation before the new behaviour takes effect.
So funktioniert KI in Magic Morning
Ein Satz: Die KI von Magic Morning läuft auf deinem iPhone über Apple Intelligence. Kein Prompt, keine Daten, keine Ausgabe erreichen unsere Server — wir haben keine.
1. Warum diese Seite existiert
Art. 50 der EU-KI-Verordnung (Verordnung (EU) 2024/1689) verlangt, dass Nutzer eines KI-Systems klar darüber informiert werden, dass sie mit einem KI-System interagieren und was es tut. Diese Seite ist diese Information für Magic Morning. Sie ist bewusst lesbar gehalten; die Rechts-Referenzen bleiben kurz.
2. Was die KI in Magic Morning tut
Die KI-Funktionen von Magic Morning erzeugen vier Kategorien von Text, alle in der App mit einem nicht entfernbaren ✦ AI-Badge gekennzeichnet:
- Weekly Recap — eine kurze narrative Zusammenfassung deiner vergangenen Woche, Sonntagabend (oder bei manuellem Öffnen) generiert.
- AI Insights — Muster-Beobachtungen über deine Gewohnheiten und Routinen, generiert ab zwei Wochen Aktivität.
- Identity-Framing-Phrasen — die "Du bist jemand, der …"-Zeile zu einer Gewohnheit.
- Habit-Trigger — Vorschläge, wann, wo oder nach welcher Aktion eine Gewohnheit ausgeführt werden soll.
3. Wo die KI läuft
Die Inferenz läuft auf deinem iPhone in Apples Foundation Models-Framework, das Teil von Apple Intelligence ist. Das Modell stellt Apple bereit, es läuft in der Neural Engine des Geräts.
Kein Prompt, kein Envelope, kein Insight und kein Recap geht an einen Server. Magic Morning hat kein Backend, keinen Proxy, keine Drittanbieter-KI-API. Schaltest du WLAN und Mobilfunk aus, funktionieren die KI-Funktionen trotzdem.
4. Was die KI lesen darf
Nur aggregierte Aktivitäten von diesem Gerät, vorbehaltlich der Pro-Kategorie-Toggles in Einstellungen → KI-Funktionen:
- Anzahl und Zeitstempel von Routine-Runs und Habit-Logs
- HealthKit-Aggregate — Durchschnitts-Schlafstunden, Workout-Minuten pro Tag, Durchschnitts-Schritte, Achtsamkeits-Minuten
- Kalender-Termin-Zeiten und Busy-Stunden
- Dein Anzeigename zur Personalisierung
Du kannst jede dieser Kategorien einzeln widerrufen. Wer "Health" widerruft, sperrt der KI den Zugriff auf jedes HealthKit-Aggregat, auch wenn die HealthKit-Berechtigung selbst weiterhin erteilt bleibt; gleiches gilt für die anderen drei Kategorien.
5. Was die KI nie liest
- HealthKit-Rohdaten (nur Aggregate)
- Kalender-Termin-Titel, Teilnehmer, Orte, Konferenz-Links, Notizen
- Kontakte, Fotos, Nachrichten, Mails, Browser-Verlauf
- Irgendetwas außerhalb von Magic Morning
6. Wann KI nicht verfügbar ist
Apple Intelligence setzt kompatible Hardware voraus (ein iPhone mit aktivierter Apple Intelligence). Auf älteren Geräten, bei deaktivierter Apple Intelligence in iOS-Einstellungen oder bevor das Modell vollständig heruntergeladen ist, sind die KI-Funktionen still deaktiviert. Du erhältst trotzdem Insights — erzeugt durch deterministische Regeln über deine Gewohnheiten und Routinen on-device, ohne Modell.
7. Wie ausschalten
Einstellungen → KI-Funktionen → KI-Funktionen → Aus. Der Schalter blendet KI-Oberflächen sofort aus. Pro-Kategorie-Toggles (Routinen, Gewohnheiten, Health, Kalender) erlauben Widerruf einzelner Datenkategorien, sodass die KI sie nicht mehr lesen kann — auch nicht on-device.
8. KI-Inhalte erkennen
Jeder KI-generierte Insight, Recap oder Vorschlag trägt ein ✦ AI-Badge. Bearbeitbare KI-Vorschläge (Identity Framing, Trigger) tragen zusätzlich ein "✦ AI suggestion"-Label, bis du sie bestätigst oder überschreibst. Das Badge ist nicht entfernbar und sichtbar, auch wenn du Theme oder Schriftgröße änderst.
9. Grenzen und Risiken der KI-Ausgabe
Das On-Device-Modell ist ein Sprachmodell. Es ist gut darin, flüssig zusammenzufassen und Muster zu benennen. Es ist nicht:
- ein medizinischer Beobachter — seine Insights sind keine Therapie, keine Diagnose, kein medizinischer Rat.
- peer-reviewed gegen veröffentlichte Studien — der Text kann gängige Habit-Forming-Heuristiken referenzieren, nicht Zitate.
- unfehlbar — kleine oder ungewöhnliche Datenmengen können zu Muster-Behauptungen führen, die in Wahrheit nicht existieren. Wir mindern das, indem wir jeden Insight unterdrücken, dessen Zahlen sich nicht auf deine tatsächlichen Daten zurückführen lassen, aber Null-Halluzinationen können wir nicht garantieren.
Behandle KI-Output als Orientierungshilfe, nicht als Anweisung. Nutzungsbedingungen Abschnitt 5 hält das verbindlich fest.
10. KI-Risiko-Klassifizierung nach EU AI Act
Die On-Device-KI von Magic Morning ist ein allgemeiner KI-Assistent für persönliche Produktivität. Sie steht nicht auf der Hochrisiko-Liste (Anhang III der EU-KI-Verordnung) und führt weder biometrische Identifikation, noch Emotionserkennung, noch Kreditwürdigkeitsprüfung oder eine andere Hochrisiko-Funktion durch. Sie unterliegt der Transparenzpflicht aus Art. 50 (die diese Seite erfüllt), nicht den Hochrisiko-Pflichten aus Kapitel III Abschnitt 2.
11. Anbieter-Verantwortung
Das Apple-Foundation-Models-Modell stellt Apple Inc. als vorgelagerter Anbieter bereit. Magic Morning ist Betreiber im Sinne von Art. 26 EU-KI-Verordnung. Wir übernehmen Verantwortung für die Art, wie das Modell geprompt wird und wie seine Ausgabe in der App dargestellt wird — einschließlich der Datenminimierungs-Entscheidungen in Abschnitt 4 und des nicht entfernbaren KI-Badges.
12. Änderungen dieser Information
Wenn wir neue KI-Funktionen hinzufügen oder den Datenfluss ins Modell ändern, aktualisieren wir diese Seite und die In-App-Erklärung, bevor das neue Verhalten in Kraft tritt.