Hark84

MusicMood Privacy Policy

Last updated: 2026-10-23

MusicMood is an Android application designed to help you understand the relationship between your music listening habits and your emotional well-being. The app operates on a principle of automated, passive analysis combined with user-controlled privacy.

Music Detection & Analysis: The app observes the songs you listen to natively on your device. Rather than relying on simple, generic database tags, basic track details are sent to Google’s Gemini AI service for analysis — extracting multi-dimensional musicological qualities (such as musical intensity, vocal delivery, tempo ranges, and core lyrical themes) that populate your personal on-device baseline.
Contextual Integration & Summarization: To transform raw listening metrics into meaningful insights, MusicMood correlates your song history with real-time context. This includes your location categories (like home, work, or gym), calendar events, weather, time of day, and current physical activity (such as driving, walking, or running), alongside physiological metrics (like heart rate and HRV-derived stress) from your wearable devices. To compile comprehensive daily summaries, anonymous contextual baselines and aggregated biometric statistics (daily heart-rate and stress ranges, hourly averages) — stripped of identity markers, never raw sensor samples — are processed by Google’s Gemini AI service to generate the natural-language summary you see in-app.
Personalized Baseline: By combining this data with specific details you provide about what different types of music mean to you, the app establishes a personalized listener profile unique to your habits.
Longitudinal Trends: Your data is structured into distinct listening sessions that aggregate into days, weeks, and months, allowing you to see how your habits shift over time.
Predictive Mood Journaling: This continuous history enables a smart mood-journaling interface. Instead of forcing you to face a blank page, the app reduces the effort of journaling by providing a predicted mood for the day alongside contextual examples of what you did and listened to, allowing you to easily validate, adjust, or reflect on your emotional trends.

The purpose of this document is to provide complete transparency regarding the specific device permissions MusicMood requests and to explain exactly how they are used to power your experience.

Stored on your device

Listening history — for each song played: title, artist, album, the time it was played, how long you listened, whether you skipped, whether you paused.
Per-song context, captured at the moment a song plays:
- Activity (still / walking / driving) from Android’s Activity Recognition API.
- Coarse location category (home / work / school / commuting / elsewhere) — derived from your GPS coordinates compared against the locations you configured in the app. MusicMood requests precise location for this: Android’s “approximate location” option deliberately fuzzes every fix to a roughly 2 km grid, which cannot distinguish your saved places or detect driving, so with approximate-only permission the location context largely stops working. Either way, only the derived category is stored with the song; raw coordinates appear on-device only in the app-private diagnostic logs described below.
- Current weather (temperature, conditions), fetched by GPS coordinates.
- Whether you’re in a calendar event, and the event’s category (work meeting, doctor’s appointment, etc.), if you’ve granted calendar permission.
- Heart rate, stress (derived from HRV), and recent sleep data, if you’ve granted Health Connect permission.
Journal entries — free-text mood reflections that you write.
AI Song Analyses — Structured musicological profiles returned by Google Gemini (including genre, vocal delivery style, tempo ranges, affective mood, and core lyrical themes) are saved locally to your history database.
Daily insights and mood summaries — text generated by Google Gemini from your recent listening + context data, plus journal entries you choose to share with the analysis.
Album art — image files cached from the playing app’s media notification.
Diagnostic logs — plain-text event log saved in app-private storage (Android/data/, unreadable by other apps), used to debug issues. A You-page “Export Logs” action copies them to Documents/MusicMood/exported-logs/ on demand.

All of the above is stored in an on-device SQLite database (musicmood.db), in cached image files, and in text log files. Your personal data — listening history, context, biometrics, journal — is never uploaded to any server we control.

The only server infrastructure MusicMood operates is a small backend (Google Firebase) with two jobs: (1) it relays AI requests so the AI API key never ships inside the app, and (2) it maintains a shared cache of song analyses. That cache stores universal facts about songs — title, artist, album, the AI’s musicological analysis, and public metadata such as release year, track duration, and popularity statistics — never your identity, your listening history, your location, your biometrics, or your journal. An analysis generated when you play a song may be served from this cache to other users (and vice versa); nothing in a cached entry indicates who triggered it. Personal-context requests (session mood summaries, daily insights) pass through the backend as a relay only and are not stored on it.

Data transmitted off your device

These services receive the listed inputs only — no other data, no personal identifiers, and no historical behavioral timelines beyond the current request. (Requests to our own backend carry the anonymous token described in its row; no third-party service receives any identifier at all.)

Service	What we send	What it returns
MusicMood backend (Google Firebase, operated by the developer)	AI analysis requests (relayed to Google Gemini as described below). Song-analysis requests and their results are stored in the shared song-analysis cache, together with the song’s public metadata (release year, duration, popularity statistics). Every backend request carries an anonymous, randomly-generated authentication token created per install plus a device-integrity attestation — both used solely to prevent abuse of the AI quota. The token is not linked to your name, email, device identifiers, or any account, and is not stored with cached song analyses.	Song analyses (from the shared cache or freshly generated) and relayed AI responses.
Google Gemini API (Google LLC, via the backend relay)	Song title, artist, album, optional release metadata (release year, genre). For session mood summaries: anonymized session-level context including aggregated activity, location category, weather, time of day, aggregated listening statistics (e.g. your recent harsh-vs-clean vocal mix), and aggregated biometric statistics (e.g. daily heart-rate/stress ranges and hourly averages) — never raw GPS coordinates, never raw heart rate samples. Calendar context is included two ways: (1) per-song/session context carries only a short keyword phrase derived from the event title plus its category (e.g. “eye doctor”, medical) — not the full title; (2) the once-daily insight and session-level mood analyses additionally include the titles of notable, non-routine calendar events verbatim — concerts, trips, holidays, and personal milestones such as weddings, anniversaries, birthdays, and memorials — within a bounded window around each event (up to 10 days before and 7 days after, extended for grief-related events), accompanied by the display name of the calendar they belong to (e.g. a subscribed “Phases of the Moon” calendar) when that name aids interpretation — because the event name itself (e.g. which band you’re seeing) is what powers those insights. Routine and recurring events (regular work meetings, weekly classes) are filtered out and never sent by title, and calendar items that are really to-do reminders are never treated as attended events. An event’s category may additionally be derived on-device from the type of venue in its location field (e.g. a movie-theater address categorizes the outing as a movie) — the address itself is never sent anywhere. For a notable event whose title can’t be categorized any other way, the daily insight may run a one-time, cached, web-grounded Gemini check asking whether the title is a currently-playing movie; that request contains the event title only. Daily insights and mood analyses can also draw on your daily mood check-ins, under the in-app controls described in “Your journal-sharing controls” below: check-in ratings (e.g. “Good”, “Rough”), and — only if you enable journal access — the text of check-in notes and journal entries.	Per-song musicological analysis (genre, mood, vocal style, lyrical themes, confidence) and session-level natural-language mood summaries.
MusicBrainz (MetaBrainz Foundation, public web service)	Song title, artist, optional album	Release year, album name, track duration
Last.fm (Audioscrobbler / CBS Interactive)	Song title, artist	Listener count and community tags. Used as a popularity anchor for AI confidence calibration; never queries your personal listening history. Only active when a Last.fm API key is configured in the build.
LrcLib (lrclib.net)	Song title, artist, optional album and duration	Synchronized lyrics
Genius (Genius Media Group)	Song title, artist (only when LrcLib has no lyrics for the song, or when you switch the lyrics source to Genius manually)	Lyrics
Open-Meteo (open-meteo.com)	Latitude/longitude	Current weather conditions
Overpass / OpenStreetMap (overpass-api.de)	Latitude/longitude	Local road speed limit (used to detect slow traffic when driving)

Use of each third-party service is subject to that service’s privacy policy. None of these services receive your name, email, permanent device identifier, full historical timelines, unshared journal entries, or raw biometric samples. Calendar data flows only to Google Gemini, only in the two limited forms described in the table above. Requests sent to Google Gemini are subject to Google’s Generative AI API terms.

Daily mood check-ins (a rating like “Good” or “Rough”, plus an optional written note) are the most personal data in the app, so their use in AI features is governed by explicit in-app controls (You page → Check-ins & Journal Options):

“Allow persona to access journal” (off by default). The master switch for written text. When off, no journal or check-in text is ever included in any AI request — only mood ratings are used. When on, check-in notes (truncated) and journal-derived snippets may be included in requests for the daily insight and for Therapist-mode mood analyses.
Home-screen mention level — None / Abstracted / Full (default: Abstracted). Controls how visibly mood analyses shown on the unlocked home screen may reference your check-in content in their generated text. Abstracted instructs the AI to keep references general (“some recent personal reflections”) and never include names or note details in what it writes; Full allows them; None additionally keeps check-in text out of home-screen analysis requests entirely. This is primarily a display-fidelity control — what data leaves your device is governed by the master switch above.
“Roast mode can see your check-in moods” (off by default). A separate opt-in for the Roast persona. When on, Roast analysis requests include your check-in mood ratings and dates only — the text of your notes is never sent to Roast at any setting.
Notifications. The daily-insight notification never contains your full insight text. It carries only a short, deliberately generic teaser (generated under a rule barring names, events, people, and places), and on a locked device with Android’s “hide sensitive content” active it shows only “Your daily insight is ready.” The full insight is only viewable inside the app.

Data retention and deletion

All on-device history and cache data is retained for as long as you keep the app installed.
Uninstalling the app removes all of the above through Android’s normal app-data cleanup.
The “Export Data” option in Settings exports a copy of the SQLite database to your Documents/MusicMood folder.
The “Clear All Data” option in Settings deletes the database and all cached files.
Individual entries (a single song play, a journal entry, an entire session) can be deleted from the relevant screen in the app.
Shared-cache song analyses on the backend contain no personal data, are not linked to any user or install, and expire on a rolling time-to-live.

We do not sell, rent, or share your data with any third party for advertising, analytics, or marketing purposes.

The third-party APIs listed above receive only the inputs specified. They are not provided with your listening history, journal entries, health metrics, or any identifier tying their request back to you.

Your rights and choices

MusicMood has no accounts, so most privacy rights are things you exercise directly on your device:

Access and portability - “Export Data” gives you a complete copy of everything the app has.
Deletion - delete individual entries in-app, use “Clear All Data”, or uninstall the app. All three act on the only copy of your personal data in existence.
Server-side data - the backend holds nothing about you except the anonymous per-install token described above; email us to have it revoked. Shared-cache song analyses contain no personal data to delete.

We do not sell personal information, and we do not share it for cross-context behavioral advertising (as those terms are defined under the California Consumer Privacy Act). If you are in a region with statutory privacy rights (such as the EU/UK GDPR or a US state privacy law) and want anything beyond the tools above, email the contact address below and we will handle it directly.

NOT collected and NOT sent anywhere

Audio recordings (the app reads metadata from media notifications; it does not record audio).
Your name, email address, phone number, or any account identifier. (The app creates one anonymous, random authentication token per install for backend abuse-prevention; it is not derived from or linked to any of these and cannot identify you.)
Device identifiers (Advertising ID, IMEI, etc.).
Calendar event details beyond what’s described above — per-song context sends only a derived keyword phrase and category, and the daily insight and mood analyses send the titles of notable non-routine events (trips, concerts, holidays, and personal milestones such as weddings, anniversaries, and memorials). Event descriptions, attendee lists, event locations, and the titles of routine/recurring events are never sent anywhere. (Event locations are read on-device solely to derive a generic venue type — “movie theater” — which is what gets sent, never the address.)
Health metrics (heart rate, sleep) beyond the anonymized aggregated ranges described above — raw samples and any identifiable clinical profile never leave your phone.
Coarse-or-finer location beyond what’s needed for weather and road speed lookups; the lat/lng we send to those specific APIs is the only location data that leaves your device.

Children’s privacy

MusicMood is not directed at children under 13. We do not knowingly collect data from children. If you are a parent and believe your child has been using MusicMood, uninstalling the app removes any locally-stored data.

Changes to this policy

If MusicMood’s data handling changes, this policy will be updated and the “Last updated” date above will change. Material changes that affect what data is collected or shared will also be disclosed in-app.

Terms of Use

MusicMood’s Terms of Use are presented for acceptance at first launch and remain available inside the app (You page) and on this site.

Contact

Maintainer: Matthew Harkless Email: support@harkcode.app

If you have questions about this policy, want to request deletion of data, or report a concern, email the address above.

MusicMood · Terms of Use · Data deletion