Let your AI listen to a song with you.

An AI can't hear. So this turns a song into pictures it can read, fifteen seconds at a time. It writes down what it notices and how it lands, before it's allowed to see what comes next. Then you play its notes back against the music.

Watch one play →Try it with your own AI

Works best if your AI uses Counterparts, human-grade memory: it remembers the songs it has heard and what you thought of them.

How it works More, with pictures →

You pick a songA song with lyrics: a YouTube link or a file of your own.
A picture of fifteen seconds of a song: the voice as bright bands with a line tracing its pitch, the instruments in rows beneath
The sound becomes picturesThe voice, its pitch, each instrument, the notes. Measured from the audio, not described by another AI.
It breaks open at 1:35, and so completely that the tool marks him as a new singer.
Your AI listens in orderOne window at a time, a note after each, locked before the next opens. No skipping ahead, so its surprise is real.

Two listens you can play

Somebody That I Used To Know, Gotye
Live on KCRW · listened to twicePlay →
 And It's Still Alright, Nathaniel Rateliff
At Pickathon · one listenPlay →

To do this with your own AI, paste this into Claude Code (Anthropic's coding tool, not the chat app). It reads the install steps, tells you what it needs and sets itself up.

What it is and what it needs, on GitHub

Free and open source. It needs a Mac with an Apple chip, about 5 GB of disk and about 15 minutes to prepare each song, and the song needs lyrics.