AI on CodeRook

What CodeRook does with your projects

Do you train AI models on the code I store here?

No. We do not train models on your projects, we do not sell them as training data, and we do not grant anyone else the right to do either. This applies to private and public projects alike — publishing a project makes it readable, which is not the same as making it material.

Does making a project public give anyone the right to train on it?

Not from us. Public means people can read and download it. Anything beyond that is governed by the licence you attach and by the switches below, not by the fact that a URL resolves without signing in.

Why should I believe that?

Because of how the service is paid for. CodeRook runs on subscriptions — storage and servers are funded by the people using them. There is no part of the business that depends on your projects being useful to anyone else, which is what makes the promise cheap for us to keep and worth something to you. It is written into our terms rather than left as a statement of intention.

The four switches

"Machines" means anything automated that reaches a project without a person driving it:

• AI crawlers and training scrapers — collecting code to train or fine-tune models.
• AI agents and coding assistants — reading or fetching a project on somebody's behalf.
• Search engine bots — indexing so people can find it.
• Mirroring and archiving tools — taking whole copies to host or preserve elsewhere.
• Scanners — dependency, licence and security tools that read code looking for particular things.
• Scripts and automated clients — anything else calling the API on a schedule rather than because somebody clicked.

A person browsing the site is not a machine, and neither is a person using the desktop application or the command line. The line is whether somebody is there.

Every project has four independent answers about machine use. All four are permissive by default, because a restriction should be something you chose rather than something you were given. You will find them in the project's settings.

Can AI read this?

Covers reading the project at all — the page, the file listings, the contents. Turning it off is also the switch that refuses model training: the project then declares ai=n, train-ai=n, carries a noai robots directive, and is named in our robots.txt so a crawler learns the answer before it fetches rather than after.

Can AI contribute to this?

Covers automated clients opening contributions against the project. It governs machine access, and says nothing at all about whether you or anyone else may use AI tools while writing code for it.

Can AI download this?

Covers taking the files — individual downloads and whole-project archives. This is about bulk collection, not about reading.

Can AI make requests?

Covers automated clients calling the project's public endpoints rather than a person browsing it.

The mistake almost everyone makes

If I turn off downloads, does that stop AI training?

No, and this is the confusion worth spending a minute on. The four switches are independent. Turning off downloading says nothing about training, and a project with downloads off can still be declaring that training is permitted.

If what you want is to refuse model training, the switch is Can AI read this? — that is the one that changes the declaration attached to every response. We have watched a careful reader get this backwards, so if it seems obvious in hindsight, it was not obvious in advance.

Using AI to build things here

Where CodeRook stands

AI is a tool. It helps developers do their work, the same as every other tool in the chain, and there is nothing nefarious about reaching for one. It is also where the industry is going, whether or not any particular platform approves — so a rule against it is not a rule about quality, it is a rule about who is allowed to publish.

Everyone is welcome here, including if you used AI. That is a deliberate position taken while others were going the other way, and it is not going to quietly change once there are enough users to make it awkward.

Can I use AI to write projects on CodeRook?

Yes, without restriction and without conditions. We do not limit AI-assisted projects, we do not have a threshold above which a project becomes unwelcome, and we do not intend to introduce one.

Will I have to declare whether I used AI?

No. Never. If you want to display how a project was built you will be able to, as a badge you switch on yourself — but it is yours to publish, not ours to require, and the absence of one means nothing at all.

The reasoning is simple. Someone who used a model for one line, or to get unstuck on a bug, did not make an AI project, and a label would misdescribe them. Compelling that disclosure is the same kind of imposition as banning the tools outright. We also do not accept the premise underneath the usual argument: authorship is not a quality signal, and plenty of AI-assisted work is better than the hand-written alternative.

Being asked to confirm you are a person

Why am I seeing a check before a download?

Because the owner of that project turned a switch on. The check is deliberately something a person passes and an unattended script does not — it is not trying to keep you out, and passing it continues whatever you were doing rather than sending you back to the start.

Does this actually stop AI?

Partly, and we would rather say so than oversell it. There are two layers. The first states the owner's answer in the places an automated client looks, which works on anything honest enough to look. The second asks for confirmation that a person is present, which stops unattended automation but not a person driving a model by hand — and that person was never the problem.

It is layers, not a wall. Someone determined to ignore both can. What it removes is the assumption that nobody minded.

If you are building an automated client

How do I use CodeRook without being throttled?

Read /robots.txt on api.coderook.com and coderook.com, and take /sitemap.xml — it lists every public project in one request, which is cheaper for you than discovering them by walking. Project and version listings are bounded and are not rate limited.

Why did I get a 429?

You were requesting per-version file routes faster than a person browses. Those URLs multiply with every version saved, so they carry a budget while the listings do not. The response tells you how long to wait; wait that long and continue.

Do you honour agent signatures?

Yes. A client presenting a Web Bot Auth signature is treated as having identified itself, rather than as having claimed something in a header anybody could copy.

What the settings are, and what they are not

Does "read" cover indexing, summarising and training?

Yes — all three. Turning reading off declares that the project is not to be used as model training or fine-tuning material, not to be collected for a summary or an answer, and not to be indexed by an AI crawler. Ordinary search indexing is treated separately and remains permitted, because being findable and being training material are different things.

Do these apply to the CLI and the API as well as the website?

Yes. The setting belongs to the project, not to the way it was reached, and the same answer is served whether the request came from a browser, the command line, or an API client.

Is this a licence, or a copyright transfer?

Neither. These are technical controls that state the owner's terms of access. They do not move any ownership, they do not replace whatever licence you have attached to the project, and they do not grant us any right to your work. What they do is remove the argument that nobody said anything — accessing a project in a way its owner has declined is use outside the permission given, and our terms treat it that way.

What can these settings not do?

They cannot promise that publicly available content is never manually copied into an AI system by a person. They cannot classify every automated client correctly, and they cannot tell a person driving a model by hand from a person reading. Most importantly, they cannot recall a copy that somebody already has — unpublishing prevents further access through CodeRook and does nothing about what has left it.

None of that makes them worthless. It makes them a statement of terms with a technical measure behind it, which is what they are, rather than a guarantee, which they are not.

If you publish AI-assisted work here

You are welcome, and you are also responsible for it — for having the rights to what you upload, for the licences of anything it incorporates, and for whether it is safe to run. A model producing the code does not change who answers for it. That is the same standard applied to anything else published here, not a special one.

Search engines

Does restricting AI hurt how the project is found?

It should not. We separate the two questions wherever the format allows it, declaring search=yes, ai-train=no as a site default, so that being findable and being training material stay distinct choices. Search crawlers are welcome here; that is the arrangement that sends people to your work.