Skip to main content

What CORPUS Is, and What It Isn't

Public beta

CORPUS is being built in the open. Some of what you read here is live, some is still design intent. Expect it to evolve.

A hand-drawn map titled 'CORPUS — Data and Value Flow'. The left side, labelled INPUT, shows individual and collective contributors feeding into a Music Library through a scoring system that measures quality, quantity, and originality. The center holds the Music Library. The right side, labelled OUTPUT, shows the library feeding rights-secure training data into generative models that power adaptive and interactive music in healthcare, automotive, robotics, art, music, and games. The bottom, labelled ECOSYSTEM, shows the financial loop: net revenue from licensed deployments flows to profit, then splits into royalties for contributors, re-investments, and owner returns, with governance rules administered by contributors themselves.
CORPUS: data and value flow. Contributions enter at the input side, are scored, and feed the corpus that trains models for downstream applications; revenue flows back to contributors and into the system. Sections of this map appear throughout the documentation to mark which part of the system a page is describing.

CORPUS works like a producers' cooperative. Musicians contribute their recordings as raw material, by explicit opt-in. CORPUS transforms that material into its own product: AI models. Licensing those models generates the revenue that flows back to the contributors whose music shaped them.

A dairy cooperative does not resell its members' milk; it makes cheese and sells the cheese. Cheese is not milk, and a trained model is not a catalog of songs. Members supply the input, the cooperative sells the transformed product, and the value returns to the members.

What CORPUS is​

  • A licensing and royalty protocol. Music is licensed on the input side by explicit opt-in. Each work is evaluated for quantity, quality, and originality relative to the existing library. Originality is rewarded because it expands the expressive range of every model trained on the corpus.
  • An open, auditable system. The scoring methodology, audit framework, and data standards are designed to be published and independently verifiable. See The Three-Layer Architecture.
  • A protected corpus. The corpus never leaves CORPUS infrastructure. Licensees license CORPUS-trained models; a federated training path for custom architectures is under development. See Access Models.
  • A dual currency. Contributors receive ongoing royalties and accumulate CRPS (Corpus Participation Rights, a lasting stake in the system their work builds).
  • Infrastructure for new markets. The markets now emerging (adaptive sound in vehicles, therapeutic music in healthcare, responsive environments in games, semantic interaction in robotics, brand-sensitive advertising, cultural and educational deployment) require music that functions as situated experience, not as product. CORPUS is built to make these markets possible. See Applications.

What CORPUS is not​

  • Not a song generator. CORPUS's focus is the licensed models and the provenance and payment infrastructure beneath them, for markets that need contextual precision and rights-clarity. Competing with consumer tools like Suno or Udio is not the primary goal, though products built on the corpus, including generative ones, remain an open option.
  • Not a streaming service. It does not sell plays or copies of your music. Value is attributed when your work enters and enriches the shared resource, not when a listener consumes it. See How Royalties Flow.
  • Not a dataset broker. It does not hand your recordings to third parties to train on. The music stays inside CORPUS infrastructure; what leaves is a trained model. See Access Models.
  • Not a buy-out. Contributing is not a one-time sale of your rights. You keep ownership, and participation is ongoing. See Ownership and Consent.
  • Not a scraper. Nothing enters CORPUS without explicit opt-in. Your music is not harvested without your knowledge or consent.
  • Not a major-label deal. A licensing agreement with a major catalog solves a legal problem but leaves the verification problem untouched: you cannot read a training set off of model weights. CORPUS treats verification as a structural property of the system, not as a contractual attestation. The full argument is in Why CORPUS.
  • Not designed to replace musicians. Models trained on CORPUS function as responsive collaborators in creative contexts and as engines for markets where music behaves rather than plays. The infrastructure is built so contributors share in what they help build.

Next: Who CORPUS Is For.