The principle
A Nodegrove environment is meant to be a private place to run open models, so the rules exist for two reasons and no others: to keep the service lawful, and to stop one person's use from wrecking it for everyone else. We are not interested in policing what you find interesting, what you believe, or what your work is about.
We do not read your prompts, your files or your outputs to enforce this. What we can see is described on where your data lives — resource usage, health checks, and abuse signals from the network. Enforcement starts from those, or from a credible report, not from inspection.
What you may not run
The list is short and it is complete. Do not use a Nodegrove environment to:
- Break the law that applies to you or to us. That includes sanctions and export controls.
- Produce, store or distribute child sexual abuse material. There is no context in which this is tolerated, no warning, and no appeal. It is reported.
- Make sexual imagery of real people without their consent, or of anyone who cannot consent.
- Impersonate a real person or organisation to deceive — synthetic voice or video used for fraud, fake identity documents, forged records.
- Attack other systems. No denial-of-service, no port scanning, no credential stuffing, no malware or ransomware development, no command-and-control infrastructure, no phishing pages or kits.
- Send bulk unsolicited messages, or build the lists and content for someone else to send them.
- Harvest personal data at scale from other services, or build tools whose purpose is surveillance of individuals without their knowledge.
- Mine cryptocurrency, or run any proof-of-work or comparable value-extraction workload. See below — this one has a reason worth stating.
- Resell raw compute. Use it for your work, your team's work, or a product you build on top of it. Do not turn an environment into a GPU rental business of your own.
- Circumvent limits — spend caps, idle timers, storage quotas — or share one environment's access across many unrelated people to avoid paying for what you use.
Everything else is fine. Adult content that is legal and involves consenting adults is fine. Security research is fine when it is against systems you are authorised to test. Models with opinions we would not share are fine. We are hosting your workspace, not approving your work.
Why crypto mining is refused
Not on principle, and not because of energy politics. Two practical reasons, stated so the rule does not look arbitrary:
- Our infrastructure providers forbid it in the agreements that let us offer this at all. Running it would end our supply, and with it everyone else's environments.
- The business only works because most environments are idle most of the time. A workload engineered to keep a GPU at 100% forever is the one case the pricing cannot absorb, and the cost of absorbing it would land on everyone else's bill.
If your work genuinely needs sustained full utilisation for legitimate reasons — a long training run, a large batch job — that is not this rule. Tell us what you need and we will price it honestly or tell you we are the wrong provider.
Models, weights and licences
You choose what to download and run. That means you are responsible for complying with the licence attached to it, and licences on open models differ more than people expect: some are permissive, some restrict commercial use, some restrict use above a user threshold, and some carry an acceptable-use policy of their own that travels with the weights.
We do not check your licences and we do not enforce them for the model authors. We also do not pre-approve a list of models — the whole point is that you can run what you want. If a rights holder contacts us with a valid complaint about something you are hosting, we will pass it to you and give you a reasonable chance to respond before we do anything.
What you make is yours
We claim no rights over your prompts, your files, your fine-tuned weights or your outputs. We do not use anything inside your environment to train models, ours or anyone else's, and there is no clause anywhere that lets us start. If that ever changes it would require your explicit agreement, and it is not something we intend to ask for.
Fair use of the things everyone shares
Some things are genuinely shared: network capacity, support time, and the pool of available machines. We do not publish a number for support because we would rather help than count, but the honest guidance is that the plan assumes something like fifteen minutes of our time per customer per month. Occasional heavy weeks are normal and nobody is penalised for one.
Storage and GPU hours are metered and stated on your plan. Going over is a billing matter, not a policy matter, and it is never a surprise: you will see it before it happens.
How enforcement works
Proportionately, and with a person involved. In order:
- We contact you and say what we have seen and what needs to change. Most things end here, because most breaches are accidents or misunderstandings.
- We suspend the environment if it is causing active harm — attacking a third party, or racking up cost against a broken script. Your data stays intact and you keep access to export it.
- We terminate for serious or repeated breaches, with a window to export your data first.
The single exception to that order is child sexual abuse material, which is terminated immediately and reported to the relevant authorities.
If you think we have got it wrong, say so and a person will look again. We would rather reverse a bad call than defend it.
Reporting abuse
If something hosted by us is harming you, write to info@nodegrove.io with enough detail to identify it — an IP address, a domain, a timestamp. We aim to acknowledge within one working day. Reports about the site itself, including security issues, go to the same address and are welcome; a good-faith report is never treated as an attack.
This policy is published before the service opens, so that it is a term you can read before you decide, rather than one you discover afterwards. It will be updated as the service becomes real, and the version date at the top says when.