Question 1

What problem does OpenRoots actually solve?

Accepted Answer

You publish something good. A company a thousand times your size builds a product on it, sells it, trains a model on it, and none of that comes back. Under MIT that is not theft. It is precisely what the licence said, and most people who chose MIT never realised that is what they were choosing. OpenRoots is the middle ground that was missing. Free for the people who cannot pay, paid by the ones who can, and training treated as the separate act it always was. Nobody is the villain in that story. The licence simply never had words for what happens now.

Question 2

Why does this matter this year rather than someday?

Accepted Answer

Agent skills, system prompts, tool definitions and protocol servers are being published right now at a volume nobody planned for. The reference collection of Model Context Protocol servers is close to ninety thousand stars, and the SDKs behind it run near half a billion downloads a month. Almost none of that work carries a licence written for what it actually is. A prompt is not source code. A model weight is not a dataset. Licences drafted for 1998 software are being stretched over artifacts they have no vocabulary for, and the gap gets filled later by whoever brings the better lawyer. ORA exists so the person who wrote the skill sets the terms first, while it is still their choice to make.

Question 3

What do I get, in one paragraph?

Accepted Answer

A licence that stays free for every person, nonprofit, school and government body at any size, and for every company under the revenue threshold, twenty million United States dollars by default, measured on trailing twelve months. No user cap, no industry restriction, no geography clause, no expiry date. Three things hold at every tier including the free one. Nobody may resell your work as a substitute for your work, training a model on it needs its own licence, and your name travels with it wherever it goes. Those three are the reason the free tier can afford to stay free.

Question 4

Which of the four do I need?

Accepted Answer

It follows from what you are releasing. Source code takes ORL. A dataset or corpus takes ORD. Model weights take ORM. Agent skills, system prompts, tool definitions, workflow templates and protocol servers take ORA. Releasing more than one kind of artifact means applying the matching instrument to each. All four share the same definitions and the same tiers, so reading one is reading all four.

Question 5

Am I in the free tier?

Accepted Answer

Almost certainly yes. The Root grant is royalty-free for every individual, nonprofit, educational institution and government body at any size, and for any company at or below the threshold on trailing twelve month revenue. Two details decide it. Revenue is measured across your whole corporate group under Section 1.4, and the threshold itself is re-indexed each January under Section 1.6. Free here means no royalty, not no terms. Attribution stays intact, training still needs a separate Compute licence, and the work still may not be resold as a substitute for itself. That is the arrangement that keeps it free for everyone below the line.

Question 6

My company is small but a larger group owns it. Which tier am I?

Accepted Answer

Canopy, from the first day. Section 1.4 defines the Legal Entity as you plus every entity that controls you, that you control, or that is under common control with you, and Section 1.5 measures revenue across that whole set. Control means more than fifty percent of the voting interest, or otherwise directing management. This is the definition adopters miss most often, so it is worth checking your cap table before assuming the free tier applies.

Question 7

Does the threshold ever move?

Accepted Answer

Yes. Section 1.6 re-indexes the Threshold Amount every first of January by the change in the United States Consumer Price Index for All Urban Consumers over the preceding calendar year. So the figure in the schedule is a base, not a permanent number. It is worth re-checking in January rather than assuming last year's figure still applies.

Question 8

What happens the day I cross the threshold?

Accepted Answer

You keep everything you built. Nothing is owed for any period before that date, no right you already exercised is withdrawn, and nothing gets renegotiated. From that day forward you owe half a percent of the revenue directly attributable to products that depend on the work, capped at two hundred and fifty thousand dollars a year however large you become. Note the two figures do different jobs. Total group revenue decides your tier, and only the attributable portion of it is charged, so the base is usually far smaller than the number that moved you up. You report it yourself within thirty days of each quarter end, in good faith. The cap is there so this can never turn into the Unity story.

Question 9

Can I negotiate the rate?

Accepted Answer

No, and that is the feature rather than the limitation. The published rate is the rate everyone gets, so you never wonder whether the company that signed before you got better terms, and you never pay a legal team to find out. A Licensor may grant a full waiver. Section 3.3 requires that waiver to be published publicly and by name, so every deviation from the published rate sits on the public record.

Question 10

Do I owe anything for internal use?

Accepted Answer

A Root Licensee owes no Section 5 royalty for internal use, and Sections 6 and 8 still apply. For a Canopy Licensee an internal operating deployment can count as Production Use under Section 1.9, so the real deployment context decides rather than the word internal. Development, testing, evaluation, research and personal non-revenue use all sit outside Production Use. AI Training Use stays separately licensed even when it happens entirely in-house.

Question 11

Can I train a model on OpenRoots-licensed work?

Accepted Answer

Yes, under a Compute licence. Training sits outside the base grant at every tier including the free one, because using a work and learning from it at scale are different acts with very different consequences for whoever made it. Section 1.10 names training, fine-tuning, distillation, retrieval augmentation, embedding generation and evaluation. Section 6 states the terms for each. Most general-purpose licences name none of them, which is why this question keeps being settled in court instead of in the licence.

Question 12

What happens if somebody trains without one?

Accepted Answer

Section 6.5 places that use outside the granted rights and preserves the Licensor's infringement remedies. Whether a particular training act infringes copyright, and which remedies are available, stay questions of applicable law and specific facts. The instrument states the Licensor's position clearly. It cannot decide in advance how a court will classify the conduct, which is one reason counsel review and jurisdiction-specific advice still matter.

Question 13

Does reading the code to learn from it count?

Accepted Answer

No. Section 6.4 protects a natural person reading, studying or learning from the work, and it protects ordinary search indexing that does not meet the training definition. The clause is aimed at machine training. Human comprehension was never the target.

Question 14

What is the provenance requirement?

Accepted Answer

If a model reproduces your work substantially word for word, the output has to say so. A Compute licensee implements reasonable technical measures so a disclosure naming the work and the licence travels with that output. It targets the exact harm alleged in the code-generation litigation, where licensed work went into training and the required notice never came out the other side. It does not try to police everything a model may have memorised, because no clause honestly can.

Question 15

Does a release ever convert to Apache, MIT, or Creative Commons automatically?

Accepted Answer

No. There is no fallback licence, no conversion date, and no clause that quietly relicenses your work once enough time has passed. The version attached to your release governs your release, permanently. A later version can be published for future releases. Applying it is your decision, never an automatic one.

Question 16

What happens if the project changes direction later?

Accepted Answer

A change ships as a new version. Published text is never edited, and every earlier version stays reachable at its own canonical address. It also avoids the contradiction of promising durable commercial protection and an automatic open-source fallback in the same document.

Question 17

Can a compliant business become a direct competitor?

Accepted Answer

Not by selling your work back to your users. Section 4 blocks selling, sublicensing, redistributing, porting, repackaging, hosting or exposing the work as a substitute product without a separate written agreement, and it binds every licensee at every tier. What anyone can still do is build something genuinely larger that uses your work inside it. That distinction is the entire point. The clause is aimed at cloning, never at adoption.

Question 18

How is a competing offering actually decided?

Accepted Answer

By the conduct listed in Section 4.1 and by whether the work itself is being commercialized as a substitute. Section 4.2 carves out internal production, a larger product with substantial independent functionality, qualifying professional services, customer-controlled deployments and qualifying marketplace listings. Those boundaries narrow the judgment without removing it. The scenario explorer separates resale, repackaging, ports, hosting, larger products, consulting, customer sites and marketplaces, so you can find the one fact that changes your answer before you pay anyone to look at it.

Question 19

Is this compatible with the GPL?

Accepted Answer

Separate aggregation works. A combined derivative work does not, because the GPL requires the whole combined work to carry GPL terms and forbids additional restrictions, and the Canopy royalty, the Compute licence and the competing-offering clause are all additional restrictions. This is a real boundary rather than an oversight. Any licence that asks the largest adopters for anything at all meets the same wall.

Question 20

Can I use it inside a closed commercial product?

Accepted Answer

Yes. Where the OpenRoots work is a component inside a larger independent product, and your product is not a substitute for the OpenRoots work itself, you ship commercially and your own source stays closed. What travels with the work is the notice, the licence version, the attribution records and the Compute restriction. Everything you wrote around it stays yours.

Question 21

Is it open source?

Accepted Answer

No, and it is not trying to be. Clause 1 of the Open Source Definition states that a licence shall not require a royalty. The Canopy tier requires one. That is a deliberate choice about what these instruments are for, not a drafting fault waiting to be corrected. No. SPDX is the standard that decides whether a licence is citable in a bill of materials, and it has never required being open source. Call OpenRoots source-available. Never call it open source, or open source later.

Question 22

How do I actually apply it?

Accepted Answer

Answer three questions in the configurator on the front page and it writes the notice for you. Paste that into your LICENSE file, complete Exhibit A and Exhibit B for your project, and publish attribution records for any third-party components. It takes minutes, costs nothing, and needs no permission, registration or notification from anyone, us included.

Question 23

Who maintains these licences?

Accepted Answer

A committee of working software developers, with a small board alongside it, and a public decision log behind every choice in the text. The guarantee is structural rather than institutional. Published versions are never edited, and every correction ships as a new version at its own permanent address. You do not have to trust us to keep the terms you adopted. The bytes are fixed, and anyone can verify them against a published SHA-256 digest.

Question 24

Has a lawyer reviewed these?

Accepted Answer

Not yet. The six questions counsel has not answered are published alongside each instrument and listed by section, so you can see exactly where the uncertainty sits before you adopt anything. Most licences carry the same open questions and simply never print them. Knowing where the soft ground is beats a confident silence every time.

Question 25

Can I fork the text?

Accepted Answer

Yes, freely. Copy it, adapt it, rename it, ship it. The words are yours to take. The one thing a fork cannot carry is the OpenRoots name, because the whole value of seeing ORL on a repository is knowing the text behind it is the same text every time. The trademark policy sets out exactly what is permitted without asking.

Question 26

What happens if the project disappears?

Accepted Answer

Your licence keeps working. The terms on the version you adopted stay the terms, and a copy you retained can be verified against its published SHA-256 digest by anyone, indefinitely. Nothing depends on us being around to act. Keep the canonical text and the digest with your release. Continuity here comes from fixed public bytes, never from a promise, a deadline, or a later relicensing event.

Question 27

Does OpenRoots collect telemetry or identify users for funding?

Accepted Answer

No. No analytics, no accounts, no advertising, no session recording. The configurator runs entirely in your browser, and the files it writes contain no telemetry of any kind. If collection is ever proposed for project infrastructure, the disclosure comes before the collection. A funding goal does not create consent to identify the people using your dependency.

27 questions, answered straight.

Why this exists

Choosing an instrument

AI training

Continuity

Compatibility and adoption

The project itself