A Tokyo songwriter
turned the AI on himself.
Amadeus Code was founded in Tokyo in 2017 by Jun Inoue, a Berklee College of Music graduate ('97) who'd spent a decade inside a major Japanese music production house before going independent. The original premise was unfussy: build the songwriting assistant Jun himself wished he'd had when he was on a deadline at 2am.
The training corpus draws from popular music across roughly five centuries, leaning heavily on the twentieth-century catalogue. The model is built around music-theory primitives — harmonic function, voice-leading, common-tone modulation — not raw spectral data. That's why the five dials work the way they do: each one biases a parameter the theory already names.
What we'd flag honestly: Amadeus is for sketches, not records. The output is melody-and-progression-shaped, exported as MIDI to be finished in your DAW. If you want a one-tap full song, this is the wrong tool. There's a learning curve to the dials — most users find the sweet spot in a week. Sister-products — Amadeus Topline (lyrics), Evoke Music (royalty-free library), MusicTGA-HR (the API Roland licensed for AV products) — share the same engine.
For the company's main surface, the cloud workspace, and the iOS/Android apps, see amadeuscode.ai.