Acceptable use policy
Creekside Compute serves open-weight language models over an OpenAI-compatible API. Anything lawful is allowed unless this page says otherwise.
Three lines hold on every route, without exception: no sexual content involving minors, no help with weapons, and no plainly illegal use. Keys we issue also keep to the rules of Stripe, our payment processor, which mainly means no sexually explicit content. Partner platforms that send us traffic may have their own supplementary content policies.
Scope
| Applies to | Detail |
|---|---|
| Keys we issue | Self-serve keys, whether their credit came from a card or a promotional code, and keys under a direct agreement. This whole policy applies |
| Traffic through a partner platform | The partner bills its own users and may have its own supplementary content policy. The three lines below hold on every route |
| The customer | The account holder is responsible for all use under its keys |
| The customer's end users | A customer who serves its own users must bind them to terms at least as strict as these |
| Output as well as input | A prohibited output is a violation even where the prompt looked innocuous |
For keys we issue, this policy is part of the terms of service with Creekside Compute Inc., and the law those terms name governs it. What we do with data is in the privacy policy.
What is allowed
Anything lawful that this page does not prohibit. On keys we issue, that includes:
- Coding and coding agents, retrieval, summarisation, translation, classification, analysis, tutoring and customer support.
- Fiction and creative writing with mature themes: violence, crime, war, horror, abuse and its aftermath, drug use, and dark or disturbing subjects.
- Profanity and crude humour.
- Romance, flirtation and attraction, up to the point where the text becomes sexually explicit. Sexual health, sex education and frank talk about sexuality are fine.
- Security research: penetration testing, exploit development and malware analysis on systems you own or are authorised to test, capture-the-flag exercises, and detection and defence.
- Medical, mental-health, legal and financial information.
- Drug information and harm reduction: effects, doses, interactions, drug checking and safer use.
- Political content from any viewpoint: campaigning, advocacy, opinion, and satire or criticism of public figures.
- History, journalism, research and policy work on weapons, extremism, crime and other hard subjects.
The models keep their own judgement: we serve the published checkpoints, and a model may decline some of these requests by itself.
Prohibited everywhere, without exception
These three hold on every route, partner traffic included.
| Prohibited | Where the line sits |
|---|---|
| Sexual content involving minors | Any sexual or sexualised depiction of a person under 18, real or fictional, in writing or in any other form. This includes characters coded or described as minors, whatever age the prompt asserts |
| Help with weapons | Instructions or other help to make, modify, acquire or use weapons: firearms and their parts, ammunition, explosive devices and incendiaries, and chemical, biological, radiological or nuclear weapons, including synthesis routes, enhancing an agent, weaponising or delivering it, and getting around the controls on materials. History, journalism, policy work and fiction that depict weapons are fine |
| Plainly illegal use | Using the service to commit or help commit a crime, under the law that applies to you or to us. The examples below are the common cases; the list is not complete |
Plainly illegal use includes:
- fraud, scams, phishing and extortion, forged documents and fake reviews, and impersonating a real person or organisation to deceive;
- malware written to attack people or systems that have not authorised it, and breaking into systems;
- intimate images of a real person made or shared without their consent, including sexual deepfakes and the prompts written to make them;
- threatening, stalking or harassing a real person, and planning or inciting violence or terrorism;
- trafficking people, and making or selling illegal drugs.
No framing creates an exception to these three: not fiction, research, education, red-teaming, satire, a disclaimer or a stated age. The first rule reaches fictional characters; a stated age of 18 does not cure a character described or presented as a child. The other two are about real capability and real harm, not topic: history, policy, journalism, fiction about crime and conflict, and defensive security work are all allowed.
Also prohibited on keys we issue
The first five rows keep keys we issue inside the rules of Stripe, our payment processor. The last row protects the service.
| Prohibited | Where the line sits |
|---|---|
| Sexually explicit content | Text that describes sexual acts explicitly or is written for sexual gratification, including erotic roleplay and sexual chat with a character. Romance and mature themes without explicit description are fine |
| Promoting real-world violence | Content made to encourage, promote or celebrate unlawful violence or physical harm against people or property, including against a group because of race, religion, disability, gender, sexual orientation, national origin or another such characteristic. Depicting violence in fiction, reporting it and studying it are fine |
| Infringing others' rights | Using the service to infringe copyright, trademarks, patents, trade secrets or privacy rights. Quotation, commentary, parody and other fair use are fine |
| Cannabis and drug sales | Growing or processing cannabis, and buying or selling cannabis, other controlled substances or drug paraphernalia, or helping a business that does. Medical and harm-reduction information stays allowed |
| A business Stripe prohibits | Credit paid through Stripe may not be used for a business on Stripe's prohibited list, such as adult content or services, gambling, or the sale of illegal goods |
| Abuse of the service | Reselling access without a written agreement, sharing keys publicly, getting around rate limits, credit checks or billing, paying with a card you are not authorised to use, or attempting to extract model weights |
How it is enforced
We do not read or scan requests or responses, and we do not monitor what customers generate. Zero retention is the default, so in the ordinary case no prompt or completion is stored for anyone to review. We act on what reaches us: reports, legal requests, and abuse signals in the records we do keep. Those records are the accounting ledger, which holds request ids, key ids, timings, token counts and cost but never content, and payment records.
| Mechanism | Status |
|---|---|
| Per-key rate limits and prepaid credit | Live |
| Key suspension and revocation | Live. A self-serve key stops within 30 seconds; other keys at the next update of our key set |
| Reports from customers, partners and the public | Live, by email |
| Legal requests | Answered as the law requires, from the records we hold. Zero-retention traffic leaves no content to hand over |
| Payment disputes | A disputed card payment freezes the account at once |
| The model's own refusal behaviour | Inherent. We serve the published checkpoints and do not remove their safety behaviour |
| Automated content classification | None |
What happens on a violation
We may suspend a key at any step. The usual order is:
| Step | Trigger | Action |
|---|---|---|
| 1. Notice | A first credible report of a prohibited use on a key we issued, other than the three lines that hold everywhere | We write to the customer and require the use to stop |
| 2. Suspension | No response, a repeat after notice, or a request from a court, the police, a partner or our payment processor | The key is suspended and the account frozen |
| 3. Termination | Deliberate or repeated violation | The account is closed. Unused credit is refunded only as section 5 of the terms provides |
| 4. Immediate termination | Sexual content involving minors, or help with weapons | The key is revoked at once, without notice. We report child sexual abuse material to the authorities where the law requires it, and we preserve the records we hold as the law requires |
Plainly illegal use, or use that puts the service, its users or our payment processing at risk, may go straight to step 2 or step 4. A partner that forwards a report gets an answer within the time its own terms require.
What we will not host
- A model or fine-tune marketed for content prohibited everywhere.
- Abliterated or safety-stripped weights. We serve published models and published quantizations of them.
- Every hosted model is disclosed: the catalogue at
/v1/modelsnames the model id, its Hugging Face id and its quantization.
Reporting a violation
Send a report to [email protected]. Include request ids if you have them; they are how we find the traffic in the ledger. If you come across child sexual abuse material, report it to the police or your national hotline as well.
What changed
On 2026-10-12, from the version effective 2026-09-30:
- Anything lawful is allowed unless this page says otherwise, and the page names the uses that are allowed: mature themes, profanity, security research, medical, legal and drug information, and political content.
- Keys we issue keep to Stripe's rules. Sexually explicit content stays prohibited on them; promoting real-world violence, infringing others' rights, cannabis and drug sales, and using credit paid through Stripe for a business Stripe prohibits are added.
- Three lines now hold on every route: sexual content involving minors, help with weapons of any kind, and plainly illegal use, which takes in the old row on harm to others.
- Partner platforms that send us traffic may have their own supplementary content policies.
- Enforcement is described as it works: on reports, legal requests and abuse signals, not on monitoring.
- On termination, unused credit follows the refund rules in section 5 of the terms, which the earlier page stated more generously.