Roland published a dated vision — an app whose promise is "AI-powered interactive communication to remove barriers in learning and performance." That app needs a musical intelligence layer that is correct every time. The API exists. The product conversation is open.
Practice continuity. Re-engagement. Personalized recommendations from real performance data. AI that removes barriers instead of replacing players. Committing that to a roadmap took conviction — and it defines exactly what the intelligence layer must do.
Wi-Fi-direct hardware, OTA updates, performance data flowing to the cloud. The pipes exist. What flows through them next is what turns connection into retention.
Neural timbre with Neutone. Composition research with Sony CSL. The pattern is established: intelligence lives beside the instrument brand. Harmony is the layer that pattern has not yet covered.
Nine out of ten people who start an instrument stop. The ones who stay are the ones who get to sound musical early — and sounding musical is a harmony problem before it is a practice problem.
The retention thesis, in two sentencesTHIRI is a chord-intelligence engine served as infrastructure: analysis, voicings, progressions, and accompaniment logic — computed from theory, the same answer every time, in milliseconds. Everything below is capability an app can call today, not a roadmap.
Feedback denominated in music, not just timing. Our shipped practice game turns chord charts into playable levels and returns étude reports — which target tones a student hit, how their line voice-led. The same engine can grade and guide practice inside any app that hears a player.
A band behind every connected instrument. The engine resolves harmony beneath a live melody and voices it for playback — so a player five lessons in sounds accompanied, not alone. Retention through sounding good early.
Performance data becomes musical insight. "You avoid the IV chord in flat keys" is a recommendation engine's dream input — and it requires deterministic analysis of what was actually played. That analysis is our core competence.
API-first, callable today. REST endpoints and an agent-native MCP surface, hosted and metered — the same shapes a connected app would call if it needed theory that does not guess.
STATUS — one API call from an app prototype. Keys at build.thiri.ai/keys.
The hardware thread is the Aerophone dossier. If you are building connected practice or performance software, write Dennison.