A SUITE OF TOOLS
AND STRATEGIES
We custom-design MIDI datasets. We hire composers and media designers. We build computational strategies that help humans discover what they didn't know they could create.
Every tool begins with composer-built MIDI datasets. Not scraped. Written by musicians on the team.
Co-Learning Factor — ratio of human decision to algorithmic suggestion, tracked every session.
Your creativity, your data, your machine. No cloud harvesting. No black-box training. Full offline capability.
State-of-the-Art Cross-Ontological method for the conversion of any media — video, image, text, emotional metadata — into structured, generative MIDI music. Not audio matching. Not style transfer. A semantic bridge between the ontology of any source medium and the ontology of musical composition — pitch, rhythm, harmonic logic, density, arc.
Feed it video, image, text or emotional metadata. Receive perfectly structured MIDI that reflects your content's narrative logic, emotional arc, and formal structure — not a random audio texture, a composed result.
The engine does not match audio. It reads the semantic structure of a source medium — its spatial composition, temporal arc, emotional weight — and maps it onto the parametric space of music: register, density, harmonic tension, rhythmic fingerprint. Patented. Unique to MUSI-CO.
We design computational strategies for human creativity — not to replace artistic judgment, but to expand the boundary of what humans can discover about their own work. When a jazz musician encounters a harmonic pattern they never generated from their own preferences. When a folk ensemble discovers variations on patterns they thought they'd exhausted. When a student composer learns that the algorithm recognized something in their voice they didn't know was there. That's what we build for. Not efficiency. Novelty. Radical experimentation. The readiness to be surprised by what you're capable of.
Custom Dataset Design. We hire real composers and media designers to build datasets from human choice, not internet scraping. Every note comes from intentional artistic decision. No shortcuts. Collaborative Learning. The algorithm doesn't decide. You do. It suggests. It reveals patterns. It expands the search space. You choose what matters. Together: novelty neither could make alone. On Your Terms. Local processing. Your data stays yours. No cloud harvesting. No black-box training. No opaque outputs. You own what you create.
Standard generative systems optimize for statistical likelihood. They learn surface patterns from massive, unlicensed datasets. MUSI-CO inverts this entirely. Our engine was designed by people who understand both music theory and machine learning. It reasons about compositional structure, not probabilistic buckets. It understands that you resolve a seventh upward instead of down for a reason — and learns that reason as a category of meaning, not as a random pattern.
Complete Control Over Sound. We generate high-fidelity MIDI, not flat audio. Total control: swap instruments, tweak velocities, rearrange notes, reprocess, iterate. Direct DAW integration with the nuance of human performance.
Semantic Sensitivity. The algorithm learns your decision-making at the level of harmonic preference, melodic contour, rhythmic fingerprint. Not statistical buckets. Categories of meaning.
Handmade Datasets from Master Composers. Small Data, Deep Intelligence. We hire composers to build intentional datasets. Every piece in our training corpus was a deliberate choice by a skilled artist. No scraping. No shortcuts.
Run Entirely On Your Hardware. Process complex generative tasks without an internet connection. Full capability, offline, on the hardware of your choice. Your creativity, your data, your machine.
Personalised orchestral soundscapes for mindfulness, focus, and restoration. Adaptive music that responds to your mental state — neuroscience meets composition.
Create the ultimate workout soundtrack with an app that generates music in real-time sync with your body's movement. Bespoke beats for bespoke performance.
Control MUSI-CO's generative engine with intuitive gestures for live performances that are never twice the same. Real-time, expressive, endlessly fresh.
Bare-bones UI preliminary release. Free to use. No implied license granted upon download. Available via the MusiCo website.
Bare-bones UI preliminary release. Free to use. No implied license granted upon download. Request access via the MusiCo website.
Classic MUSI-CO Compositions. Historical generative recordings.
The flagship engine. Feed it video, image, or emotional metadata — receive perfectly structured MIDI that reflects your content's narrative logic.
Open Demo Portal → Live PlatformLive Human–AI music production platform. Generative looping from intentional MIDI models: evolving, non-repetitive material for scoring, games, installations and sonification. Text-to-MIDI integration in the Live Action via GLoope.
PerformanceReal-time AI music generation for live performance. Low-latency local processing — no internet dependency, no stage failures. Responsive musical accompaniment.
Open LiveAI → Listening & CurationA framework for understanding what your music does. Analysis, pattern recognition, compositional insight — turning generation into learning.
Open Listen → Research · MeasurementHuman-AI Creative Understanding: our research framework for measuring musical novelty — how far a human–machine co-composition departs from what either could have produced alone, and how that distance can be evaluated with rigour rather than taste.
Open HAICU → Archive · AudioHistorical generative compositions, early experiments, and demo recordings from MUSI-CO's development journey — available freely on SoundCloud.
Listen on SoundCloud →Our strategic partner for Ethical AI governance, AI Act compliance, and GDPR alignment. AssistiveLawTech provides the legal intelligence that ensures every MUSI-CO system is designed to perform and to comply — transparently, rigorously, by design.
assistivelawtech.com →MUSI-CO actively participates in the Super AI Studio open-source community — contributing to and integrating state-of-the-art LLM and SLM architectures into our music generation pipeline. Strategic consulting for European PMIs. EU-radically compliant. Vendor-agnostic technical architecture.
superaistudio.com →An ongoing research collaboration spanning nearly a decade of generative MIDI development. From early rhythm models (2017) through VAE/GAN interpolation (2018) and CONLON at ISMIR (2020), to forthcoming diffusion-based approaches (April 2026).
Early explorations of generative models for EDM drum patterns.
Workshop Paper →VAE/GAN-based generation with smooth genre transitions.
arXiv Paper →Lossless format, Wasserstein autoencoders, latent-space trajectory generation.
CONLON Paper →The forthcoming paper replaces autoencoder latent-space interpolation with diffusion models. This shift offers richer modelling of the full MIDI distribution, better handling of polyphonic multi-instrument structure, and more controllable conditioning across style, genre, and harmonic context.
An ongoing interdisciplinary collaboration applying MUSI-CO's core technology building blocks beyond music — into advanced simulation strategies for assistive urban mobility. Developing adaptive multimedia storytelling systems for accessible navigation and wayfinding experiences in complex urban environments.
Process complex generative tasks entirely without an internet connection. Full capability, offline, on your hardware of choice.
Your compositions and media assets never touch our servers. What you create is yours — completely, legally, permanently.
In collaboration with AssistiveLawTech, MUSI-CO's architecture is designed for full compliance with the EU AI Act and GDPR — privacy and ethics baked in from day one.
Our models were never trained on your work without consent. Every dataset is licensed, human-composed, and fully documented. Transparency is a foundation, not a feature.
MUSI-CO was founded on the belief that AI should be a digital luthier — a maker of fine tools for fine artists. We reject the "one-click" commodification of music. Instead, we provide a sophisticated framework where human composition and algorithmic precision coexist as equals.
"We custom-hire composers and media designers to build datasets. We embed ethical choice into every layer of code. We believe creativity expands when humans and algorithms collaborate as partners, not when one dominates the other."
Our team unites composers, music theorists, engineers, and strategists under one roof — a rare combination that produces tools no pure-technology company could build. Every model we ship has been listened to, critiqued, and refined by people who have spent decades studying what makes music move people. We are not anti-AI. We are pro-craft. The distinction matters enormously.
A suite of open experiments in co-learning — the process by which a human and an algorithm build shared competence through interaction. Nothing here is finished. Authorship is tracked, novelty is measured, the composer stays in the loop. Each tool is an independent fork; every fork links directly to a live instrument.
Six profiles — theatre and cinema directors, choreographers, media artists, galleries and collectors. Maps a practice from its conscious and submerged sources, verifies the rights chain, and reads formal divergence against a public, versioned baseline. Deterministic local scoring; AVA narration suspended; the material never leaves the browser. 6 profiles · 2 polarities: mirror vs lens · P0–P8 mapping · IP matrix · no-training default · per-profile agents.
Draw the sound. Four stem layers emerge by quantity; the tempo ruler sets BPM by horizontal position and hue. Generate re-rolls all layers chromatically. Rec: audio · MIDI · audio-video.
Camera + voice. One finger up triggers a cell; fist generates; two fingers mute. Both open palms set BPM. Palm close/open pulses a jellyfish filter. Thumb-up shaken re-randomises. Genre select changes the library folder.
Five string-quartet movements cut into four instrument cells. Each plays the score linearly. GLITCH axis slides from pure form to granular synthesis. A text prompt steers it all; a drum cell adds drumming underneath.
Record your own gestures on camera, name them, bind them to any control or sound-processing action. Every detected gesture is re-mappable: trigger, generate, mute, solo, loop, reverse, transpose, filter, tempo, reverb, delay. Saved in browser, runs on-device.
Five Le Corbusier fundamentals — pilotis, free plan, free façade, ribbon windows, roof garden — drawn as radical primary volumes in 3D, each bound to a macro. Step INSIDE (IMMERSIVE) for a first-person walk; the song deconstructs into granular, time-stretched, spatialised sound.
Box the grid. Jab depth (palm span) maps to sound family — AIR / BODY / IMPACT. Lateral position picks the cell. Hysteresis and smoothing prevent jitter. Live readout shows depth, bin, cell and family per hit.
Symmetry answers. Repetition rewards. DRONE is the safe default — rest, go steady or move asymmetrically and it stays quiet. Move symmetrically: consistent tonal reaction. Repeat: analogue arpeggios and harmony grow out of the drone. Analogue synths only.
Built on the SYNTAGMA DAW stack. Four cells, per-cell routing. Slow fingers = sustained legato; fast fingers = staccato arpeggios. Waves → smooth ±10 BPM per cell. Swipe → reverb, sustain, pedal per hand. Piano / electric piano / vibraphone / xylophone.
Open platform for multimedia stories about Italy — beginning in Scicli, Sicily. Pin a spot on OSM, upload photos, video, audio, PDFs (max 10 MB). Local IndexedDB archive. AI topic agent queued for future processing.
Shared console for collective prompt texting during live and online events. The audience writes musical instructions; an LLM filter panel weighs safety, relevance, playability and freshness. Accepted prompts become a synchronised score for live musicians, or a spatial soundscape for improvisers. Rendered live in WebAudio; no text leaves the browser.
Full interactive demo suite for all MUSI-CO platforms and services.
Open Demo Portal →Historical generative compositions, demos, and early MUSI-CO recordings.
Listen on SoundCloud →Interviews, partnerships, speaking engagements, and collaborations.
info@musi-co.com →
Preliminary release • Free to use
VST Plugin • Mac
Preliminary release • Free to use
VST Plugin • Mac
