Music has always been social: a Harvard-led study ↗ found song in every one of the 315 societies it examined, tied to the moments people share. Streaming solved access to music. Sharing it with the people around you is still catching up. GRAI, an AI music company, is building a social streaming service for a younger generation, where people discover music through each other, listen together, and have fun with it.

To accomplish this goal, GRAI needed a catalogue it could stream from day one, rich enough to recommend from the music itself, and licensing in place so every play is paid for. And because GRAI sees the future of streaming as co-creative as well as social, it is building its own music model, which needs data licensed specifically to learn from. GRAI came to MassiveMusic for both.

Challenge

GRAI needed a catalogue large enough to build on, licensed for the intended playback uses, with metadata to power recommendations, and musically rich enough to capture the nuances of style and structure. Most catalogues only offer thin descriptions instead: a genre, mood, or tempo. For someone who has just started, with no listening history yet and very little known about their taste, thin isn't enough. The service has to work from the music itself. Training data raised a separate bar: streaming permission alone does not establish permission for AI training, so GRAI needed data cleared for exactly that use.

Two friends laughing as they share a pair of earphones, music moving between people rather than playing for one

Approach

First, we started with the rights infrastructure. MassiveMusic provided access to a large, rightsholder-approved foundation, with clear governance around permitted use built in from the outset. That infrastructure covered two uses, handled separately: a dedicated, opt-in dataset cleared specifically for AI training, and a pathway to bring eligible catalogue onto GRAI for streaming.

The second layer was metadata. Raw audio alone tells a recommendation engine what a track sounds like, not what a listener will want to hear next. Audio paired with enhanced metadata, covering musical attributes including genre, mood, energy, structure and beyond, gives the system enough context to make an educated call from a listener's very first day.

The third layer was delivery. A recommendation only matters if the track arrives cleanly when it's called, so audio streams to GRAI's users directly from MassiveMusic's own media delivery platform, letting GRAI focus on the experience rather than the infrastructure underneath it.

Group of laughing friends in colourful outfits, the younger generation GRAI's social streaming service is made for

Solution

MassiveMusic supplied end-to-end music infrastructure across two distinct use cases: catalogue delivery with enhanced metadata to power GRAI’s recommendation engine from day one, and a dedicated dataset licensed specifically for AI training. In all, MassiveMusic provided GRAI with:

  • Rights-managed delivery: delivery controls based on supplied rightsholder permissions, supporting access to content approved for the relevant use and territory

  • Catalogue licensing support: a route to participating independent and distributor catalogues through MassiveMusic's established rightsholder agreements, supporting quicker time to market

  • Seamless catalogue access: a single normalised metadata feed designed to scale to the catalogue licensed for GRAI through MassiveMusic, with cleared titles delivered to GRAI from suppliers within minutes.

  • Enhanced metadata: our full set of enhanced metadata, with 30-plus metadata dimensions, powering GRAI's search and recommendation engine

  • Reliable streaming: audio streamed directly from MassiveMusic's media delivery platform to GRAI's end users, allowing GRAI to reduce the storage costs and overhead of managing catalogue delivery

  • Reporting: plays on GRAI reported back to the licensors behind the music

  • Dedicated training data: an opt-in, rightsholder-approved dataset licensed specifically for AI training, spanning full-length tracks across genres, decades and regions

Young guitarist with a friend in a poster-lined room, hinting at GRAI's co-creative vision for music

Results

The first builds of GRAI are already in testers' hands, with MassiveMusic supplying the test catalogue, the metadata and the streaming infrastructure. It's an operational base GRAI can keep building on.

The same rights-first approach carries into what comes next: GRAI is building towards letting people put something of their own into the songs they love, with artists, songwriters and other rightsholders asked and paid. In August 2026, it became one of the first two companies to license the patents of Music IP Holdings, which cover technology for using recorded music with AI, from creation and authorisation to distribution and payment.

"In engineering, you don't want to build on shaky ground and hope to fix the architecture under pressure. For us, establishing 'rights before code' from day one was an operational necessity. When music moves between people, the rights have to move with it. Partnering with MassiveMusic gave us that foundation right out of the gate, letting us build with total clarity and focus on the product." — Dzmitry Kamarouski, CTO and co-founder of GRAI.

fallback
LQIP