The story
I built Magpie Standard to answer one question: when an AI company trains on a creator's work, who said yes, and who got paid?
That question had nothing behind it. Collecting societies were built for radio play and mechanical royalties, not for machine-readable consent, work by work. So creators were left watching their work turn up in AI models with no way to permit it, price it, or refuse it. Magpie Standard was the answer - a white-label platform that let collecting societies, unions and trade bodies offer consent-based AI licensing under their own brand. Every one of the seven building blocks the UK government's ACCCT framework said this needed, built and working.
That was the start. It was not the end, because building the answer taught me something I had not expected.
AI licensing was not a problem of its own. It was the newest symptom of an older one. The question underneath "can AI train on this?" is the same question creators have never been able to answer about any use of their work: where is it, who is using it, on what terms, and was I paid what those terms say? AI was simply the use that made the gap impossible to ignore. And the machinery that answers a training request is the machinery that answers a sync, a stream, a licence, a reprint. Register the work once. Set the terms - per work, per use. License it. Then check every use against what you agreed. AI is one lane out of more than thirty, and it gets a yes, a no, or a price, like every other use.
So I built that. Not an AI-licensing tool with a bigger marketing page - the rights platform for creative work, with AI as one lane inside it. It is called Parley, after the meeting where two sides come to agree terms before anything happens.
Consent. Your terms, per work and per use, served machine-readably, so a compliant system gets your answer before it acts, not after.
Credit. Everyone who made it, on the record: every contributor, their role, and their consent captured before a work is licensable. Declared human-made, and said so.
Control. Register once, fingerprinted and mapped to every identifier your work already carries. Allow it, block it, or ask-me-first, on every use. Say no to everything and the no is still logged and still holds.
Compensation. Your price on every use, from sourced benchmarks, not guesses. And the receipts: who read your terms, snapshotted, time-stamped and hashed; a certificate on every completed licence with the checks attached; every statement you upload checked line by line against what you agreed - authorised, underpaid, or unauthorised. They sample. We reconcile.
Not a claim that we can see inside a model. Proof of what was served, what was read, and what was agreed.
Six patent-pending methods. Built on open standards. And where a right is covered we cite the source; where it is not, we show you the gap instead. We would rather show you a gap than sell you a guess.
Magpie Standard did not go away. It became the ground Parley stands on - the standard and the methods underneath the platform, and the name the rights organisations already know when they come to the table.