Private beta

Cued
Music

your taste, on cue

01 · The bet

A research project, run on one listener at a time.

The field pools the behavior of millions of listeners and generates from the sequences. Cued bets the other way: one listener, no crowd data, and a signal no platform logs — the listener's own graded verdicts.

request→ LLM routes→ graph retrieves→ audio ranks→ pack plays→ graded verdicts

↻ verdicts tune the taste model for the next pack · no crowd data at any step

136,651tracks embedded — heard before served
8,012serves · 2,940 distinct artists
73%served from outside the listener's library
1,630graded accept / reject verdicts

Production log · 4 listeners · 726 sessions · read 2026-08-04

H1validated

Audio-ranked discovery finds music you love outside your history.

H2in test

Taste separates into moment-conditioned dimensions — 54 live now.

H3open

Graded verdicts improve the next pack — 1,630 banked.

02 · Conditional inference

A request is not a playlist. It is a hypothesis about your taste.

Ask for Disney, kids music, the 90s, or a phrase only you would use. An AI system interprets the language; the recommender tests that interpretation against your library before it serves a track.

Language
era signal / “90s”
Personal prior
artists, albums, and sound already present in your library
Ranking
conditional taste vector in 512-dimensional audio space
Retrieval
artist similarity constrained by scene and culture
Memory
saved as “your 90s” and refined by feedback

03 · Exploration objective

Discovery should widen taste without erasing identity.

Cued moves from music you know, through the scenes and artists around it, toward music you have not heard. The objective is new-to-you × good—not another route back to the same familiar hits.

score(candidate) = taste fit × moment fit × unfamiliarity

Deep catalog: beloved albums, b-sides, and buried favorites establish the personal prior.

Retrieve artist graph + scene boundary
Rank 512-dimensional audio fit
Remember served songs do not re-enter the same journey

04 · The instrument

Return to the music that formed you. Leave with something new.

Start from a Moment, an era, one song, or a beloved album. Cued plays five full tracks through Apple Music, keeps each journey distinct, and learns from every keep, skip, and play-through.

InputMoment · era · song · album
Outputfive uninterrupted full tracks
Statesaved taste × moment memory
Modedeep catalog + new-to-you discovery
Nowthe H3 verdict — do graded sittings measurably improve the next pack?
Nextuninterrupted packs on a locked phone
Thenpublish the verdicts here — including the negative ones
Private beta — request an invite → iPhone · Apple Music subscription · free beta

Cued reads only the Apple Music library you connect. Full tracks play through your account. Privacy policy.