Library AI: Playlists & Questions
Library AI is a conversational assistant for your entire music collection. Ask it to build playlists, answer questions about your listening habits, or tell you about artists in your library. It uses Apple’s on-device Foundation Models (iOS 26+) with real-time tool calling into your local database — no internet required after the initial artist database fetch.
Requires Apple Intelligence — iPhone 15 Pro or later, or any M-chip iPad, running iOS 26+, with Apple Intelligence enabled in Settings.
Opening Library AI
Tap the sparkle icon (✨) in the Library screen header.
[SCREENSHOT: Library AI modal open — header with track/artist count, suggested action pills]
The modal shows your library size (tracks and artists) in the header, suggested action chips, and a text input at the bottom.
Suggested Actions
Quick-tap actions at the top of the home state:
- Make me a chill evening playlist
- Build a high-energy workout mix
- What should I listen to next?
- Analyze N tracks (appears when unanalyzed tracks exist — see Analyze Your Library)
Tap any to send immediately.
Playlist Generation
Library AI can build playlists from almost any description. Before selecting artists, the model always calls tools to search your actual library — it will never suggest an artist you don’t own.
Examples of things you can ask:
- “Make me a rainy day playlist”
- “Build a workout mix from my library”
- “Give me something with the same vibe as my jazz collection”
- “Anything Dave Grohl was in”
- “All the grunge I have”
- “Play my most-listened artists”
How it works: depending on your request, the model calls one or more of these tools before building the playlist:
| Tool called | When | What it does |
|---|---|---|
searchLibrary | General mood/vibe requests | Searches your library artists by keyword |
searchArtistsByGenre | Genre-specific requests (“grunge”, “shoegaze”) | Matches artists by their verified genre tags |
searchArtistsByMember | Member-based requests (“Dave Grohl”, “Kim Gordon”) | Finds artists featuring that person |
rankArtistsByListening | Favourite/most-played requests | Ranks by your actual play count history |
The model is instructed to only include exact artist names returned by its tool calls. It cannot invent or hallucinate library entries.
When a playlist is generated, the result shows the playlist name, track count, and the artists included. Tap ▶ Play Now to start immediately. The playlist is saved to your Playlists tab.
[SCREENSHOT: Playlist result card — name, artist chips, Play Now button]
Mood-Based Playlists (Requires Library Analysis)
If you’ve run music theory analysis on your library (see Analyze Your Library), Library AI can build playlists using actual audio data — not just genre tags. It calls searchTracksByMood to filter by the mood, key, and BPM values detected in the audio:
- “Give me something melancholic and minor key”
- “High energy tracks above 130 BPM”
- “Peaceful, slow — nothing aggressive”
These playlists are built track-by-track (not artist-by-artist) and can pull different songs from the same artist if some match and others don’t.
Library Questions
Ask about your collection or listening habits in plain language:
- “What artist do I listen to most?”
- “When do I usually listen to music?”
- “What genres do I have the most of?”
- “Tell me about Radiohead”
- “What albums do I have by Nick Cave?”
For personal statistics questions, the model calls getListeningAnalytics to read your actual play counts, skip rates, and listening time — it never estimates or guesses these numbers.
For artist questions, the model calls getArtistFacts to read from the verified artist database. If an artist isn’t in the database, it will say so rather than guess from its training memory.
Follow-Up Questions
Every response includes 2–3 follow-up question chips relevant to what you just asked. Tap any to continue the conversation without retyping.
[SCREENSHOT: Follow-up pills below a response]
Action Chips
Responses may include action chips for direct playback:
- ▶ Artist name — tap to navigate to that artist
- ▶ Album title — tap to start playing the album
- ▤ Playlist description — tap to generate a playlist based on the suggestion
Artist Card
When a response is about a specific artist (and their facts are in the database), a compact Artist Card appears below the text showing verified genres, origin, formation year, and members.
What the AI Will and Won’t Say
The model is strictly grounded in your library data and the verified artist database. It is explicitly instructed not to fill in factual gaps from its training memory. For listening statistics especially — play counts, skip rates, listening hours, peak times — it always reads from your actual local data. These numbers are not estimated.
For artists not in the database, or statistics that don’t exist yet, it will say it doesn’t have the information.
Starting Over
Tap Start over below any response to clear the conversation and return to the home state with suggested actions.