The secret isnot the model.

Reading, not scraping

The first thing most tools get wrong is the reading. They run a contract through optical character recognition, turn it into a stream of loose text, and lose the layout that carried half the meaning.

WALDHORN.AI hands the document to a multimodal model that sees the actual page: the columns of a royalty schedule, the indent of a sub-clause, the signature block, the initials in the margin.1 A garbled scan does not become garbled advice, because nothing was flattened into plain text first. The engine reads the way a person reads, then works from what it saw.

Taught by the real world

Ask what a legal AI was taught by. The honest answer, for most of them, is the internet.

Ours is different. The detection layer began on paper: hundreds of real-world agreements, the kind that actually get signed, spanning more than thirty kinds of deal from recording and publishing to touring, producer, brand, and sync. Working entertainment lawyers read them with us, deal by deal, and marked what they would flag and why. That analysis was distilled, by hand, into the two artifacts that make the engine what it is: a detection playbook that encodes how counsel reads each kind of agreement, and a graded clause library, red, amber, and green, kept current as the business changes.2

The frontier model underneath is a component, and we swap it as better ones arrive. The distilled layer is ours, and it is the part no one can shortcut: you cannot scrape it from the internet, because it was never on the internet.

A specialist reads differently

Music contracts have their own language, and it is exactly where a general tool slips. Ask a generic chatbot about a clause and it will summarize the words. It will not tell you that the summary is where your masters quietly change hands.

The difference is not intelligence. It is context, held on purpose. Move down the rows below and watch the same question answered two ways.

Graded, and anchored to the page

Every finding carries a grade, so you can read the risk of a contract before you read a word of it:

• Red for a deal-breaking term, the kind that follows a career for decades.
• Amber for a material risk worth negotiating before you sign.
• Green for terms that are standard, or already in your favor.

A grade you cannot check is just an opinion. So a second pass finds the exact words each finding refers to and pins the note to that span on the page, the way a lawyer marks up a hard copy. If the engine cannot locate the language cleanly, it says so rather than guessing. You can export the marked-up copy, and the reasoning travels with it.

It writes for 2026, not 2016

Reading is defense. The same engine also writes.

Build mode drafts an agreement from a guided intake, using the distilled knowledge in reverse: instead of detecting the traps, it writes the protections in. And it drafts for the world the deal will actually live in. An agreement drafted here speaks to AI training and voice cloning, to streaming-era royalties, and to reversion terms with real dates, because a contract written for 2016 quietly fails its signer in 2026.3

More than a reader

Review and drafting sit inside a workspace that closes the whole deal: negotiation with tracked redlines, e-signature with sealed and independently verifiable envelopes, a forensic royalty audit that checks statements against the deal that produced them, and a rights vault that holds what you signed and what you own. Every step reads from the same distilled knowledge, which is why the loop holds together. The full map is in The Category.

And it all happens on a human timescale. A full review streams onto the page in minutes, clause by clause, while you watch the contract get read.

Where a human stays in the loop

A tool that speaks with the authority of counsel has to earn it. Four commitments are built into the workspace rather than bolted on:

• Conservative under uncertainty. When the answer depends on jurisdiction or facts we cannot see, the workspace says so and points to a human, instead of inventing certainty.
• No invented law. It does not cite statutes or cases that do not exist. Where it is unsure of an authority, it names none.
• Your documents are yours. Uploads are not used to train models unless you explicitly opt in, and that consent is revocable.
• A tool, not a lawyer. WALDHORN.AI is not a law firm and forms no attorney-client relationship. For the decisions that matter, it makes you ready for counsel, not a substitute for it.

Those are not disclaimers hidden in a footer. They are the reason a non-lawyer can trust what they read here.4