/TERMSTerms of Service

Terms of Service

Plain terms for a product that is still young: what you may do with it, what the credits are, how billing works, and what we do and do not promise. Where something is genuinely uncertain — availability at this stage, what a language model will produce — it says so rather than burying it.

Last updated
Version
terms@2026-09-27
History
Every revision is a commit in this page’s repository.

/TOS_01The agreement

The agreement

These terms are the agreement between you and Super Protocol Corp., 65 Oceana Dr E, Apt 2F, Brooklyn, NY 11235, United States (“we”, “us”) for the use of Confidential Router: the console, the OpenAI-compatible API, and this page. You accept them when you create an account, and the version you accepted is recorded with the date you accepted it.

The Privacy Policy is part of this agreement and describes what we process. If the two documents ever appear to disagree about what happens to your data, the Privacy Policy is the one that describes the system.

/TOS_02Eligibility and your account

Eligibility and your account

  • You must be at least 18 and able to enter a contract. If you are signing up for a company, you must be authorised to bind it.
  • One person, one account. Creating accounts in bulk — in particular to collect more than one credit grant — is a breach of these terms, and the grant logic is built to refuse it.
  • Keep your sign-in details and your API keys to yourself. Anything done with your keys is charged to your balance, and we cannot tell the difference between you and someone holding your key.
  • Give us an email address that reaches you. It is how we send sign-in links, receipts and anything we owe you notice of.

/TOS_03API keys

API keys

A key is shown to you once and stored only as a hash: we cannot recover it, so if you lose it, create another and revoke the old one. You can scope a key to particular models, cap what it may spend, and give it an expiry. Revoking a key stops it immediately; the usage it already metered stays on your account, because it was already served.

/TOS_04The service, and what it does not promise

The service, and what it does not promise

We provide an OpenAI-compatible inference endpoint served by models running in confidential hardware, a console to manage keys and credits, and published attestation evidence you can verify with our open-source Gatekeeper.

  • The model catalogue changes. We may add, remove or replace models, and prices can change. A change to a model’s price applies to requests made after it; requests already served keep the price they were metered at.
  • There is no service level agreement at this stage. We aim for the endpoint to be available and fast, and we do not commit to an uptime figure, a latency figure, or a support response time. If that is a requirement for you, talk to us before you build on it rather than after.
  • Verification is yours to run. The evidence is published whether or not you check it. If you run the Gatekeeper and pin a deployment, a redeployment we make will be refused by your own policy until you review it — that is the design working, not an outage.
  • We may change how it works. Features can be added, altered or withdrawn. If we withdraw something you are relying on, we will say so by email before it happens where we reasonably can.

/TOS_05Model outputs

Model outputs

The models we serve are open-weight models we did not train. We do not warrant that their output is accurate, complete, current, lawful, non-infringing or fit for any purpose, and output is generated, not reviewed by us — indeed we cannot review it, because we do not see it. Treat it as a draft from a capable but unreliable colleague.

You are responsible for what you do with output: check it before you rely on it, and do not use it as a substitute for professional advice, as the sole basis of a decision about a person’s rights, employment, credit, health or safety, or anywhere a mistake is expensive. Two users sending the same prompt may get different answers, and the same prompt may answer differently tomorrow.

/TOS_06Acceptable use

Acceptable use

Do not use the service to:

  • break the law, or help anyone else break it;
  • generate material that sexually exploits children, incites violence, or is designed to harass or defraud a specific person;
  • build weapons, malware, or a campaign of unsolicited messages;
  • attack the service or the cluster it runs in — probing the isolation boundary, attempting to reach another tenant’s traffic, denial of service, or working around rate limits and spend caps;
  • resell raw access under another name while presenting it as your own confidential infrastructure, or represent that we verified, endorsed or reviewed what you built;
  • collect credit grants you are not entitled to: sharing invitation codes you were not sent, creating accounts in bulk, or automating sign-ups.

Confidentiality is not impunity. We cannot read your prompts, and we do not intend to be able to — so we enforce this clause from what we can see: metering, error rates, abuse reports, and the fingerprints in §7. The remedy for a breach is suspension, not inspection.

You keep us harmless from third-party claims that arise from your own breach of this clause. Nothing here makes you responsible for the way the service itself works.

/TOS_07Credits

Credits

Everything on the service is paid for in credits, denominated in US dollars, and consumed by metered usage: each request debits its own cost, computed from the model’s published prices and the tokens in and out. Credits can arrive in two ways — granted, or bought — and at launch only the first one is switched on.

Granted credits

  • They have no cash value. They are a courtesy grant of service, not money, not a deposit, and not a claim on us.
  • They are non-refundable, non-redeemable and non-transferable. They cannot be paid out, exchanged, sold, redeemed for money or anything else, or moved to another account.
  • One grant per account. An invitation code credits exactly once, to the first account that redeems it, and an account can hold one invitation grant — not one per campaign. The feedback grant, where offered, is a separate one-off on the same terms.
  • An invitation code can expire or be withdrawn before it is redeemed; if it carries an expiry, the invitation says so. Credits already in your balance do not expire.
  • We may withhold or reverse a grant obtained by breaking §6 — bulk sign-ups, or a code that was not sent to you.

There is nothing to buy yet, and so nothing to refund

There are no real-money purchases on the service today. Paid top-ups are not enabled: every balance in existence is a grant — an invitation grant, or the feedback grant where it is offered — no card is charged, and no payment is taken. So the refund question has a factual answer rather than a policy one: granted credits are not money, carry no cash value, and are not refundable or redeemable.

§8 describes how paid top-ups work for when they are switched on, because terms you accepted should already say what happens rather than arriving with the first invoice. If we switch them on, this section and §8 are rewritten — with a refund position in them — before any payment is taken, and that rewrite is a dated revision of this document, notified under §13.

Usage is debited from whatever is in the balance; granted and purchased credits are one pool, so a grant is spent before any money is, by simply being there first. A request is admitted while your balance is positive and the key’s own spend cap is not exhausted. A single response may take the balance slightly negative — the length of a completion is not known before it is generated — and further requests are refused until the balance is positive again. Prices are per million tokens, in and out, published beside each model, and frozen on each request’s record at the price it was metered with.

/TOS_08Billing

Billing

None of this is switched on today — see §7. Everything below describes how paying for credits works once paid top-ups are enabled; until they are, no part of this section applies to anyone, and what is true now about metering and admission is in §7 instead.

  • Top-ups are card payments through Stripe Checkout, with a minimum amount shown at checkout. Your card details are entered on Stripe’s page and never reach us.
  • Automatic top-up is opt-in. If you enable it, we charge the card saved at your last checkout when your balance falls below the threshold you chose, for the amount you chose, at most once per hour.
  • A balance you have paid for is prepaid: it does not expire, it is not transferable between accounts, and it is consumed at the prices published when each request was served.
  • Taxes: amounts are exclusive of any VAT or sales tax, which is added at checkout where it applies.

/TOS_09Your content stays yours

Your content stays yours

You keep every right you have in what you send and in what the model returns to you, as far as it is ours to leave with you. We take no licence in your prompts or completions. We do not train on them, review them, or share them — none of which we could do, since we do not store them and cannot read enclave memory. This clause is here to say so in the document where a licence grant would normally be hiding.

Feedback you choose to send us — through the feedback form, an email, an issue — we may use to improve the product, without obligation and without it becoming confidential. That is the one thing in this section we do take.

/TOS_10Suspension and closing an account

Suspension and closing an account

We may suspend a key, a workspace or an account — immediately where the risk warrants it — for a breach of §6, for abuse of grants, for non-payment, or where we must in order to keep the service running for everybody else. We will tell you why, and restore access if the cause is resolved.

You can stop using the service at any time. To have the account and its data deleted, write to privacy@superprotocol.com — the Privacy Policy describes exactly what is deleted and what is kept for accounting. Granted credits end with the account, and there is no paid balance to settle — §7.

/TOS_11Warranties

Warranties

The service is provided as it is and as it is available. Beyond what the law will not let us exclude, and beyond the specific factual claims we make about attestation — which are verifiable and which we stand behind — we disclaim implied warranties of merchantability, fitness for a particular purpose and non-infringement.

One promise we do make plainly, because the product is built on it: we do not store your prompts or completions, and the deployment publishes evidence of the code that handles them. If that ever stops being true, the evidence will say so before we do.

/TOS_12Limitation of liability

Limitation of liability

To the extent the law permits, neither party is liable for indirect, incidental, special or consequential loss, for lost profit, revenue, goodwill or anticipated saving, or for loss or corruption of data that was not caused by that party’s own breach.

Our total liability arising out of or relating to this agreement is limited to the greater of the amount you actually paid us in the three months before the claim, and one hundred US dollars. Granted credits are not an amount paid and do not count towards that cap. Nothing here excludes liability for death or personal injury caused by negligence, for fraud, or for anything else that cannot lawfully be excluded.

/TOS_13Changes

Changes

We may change these terms. The date at the top is when the current version took effect, every revision is a commit in this page’s repository, and material changes are emailed to account holders before they take effect. Continuing to use the service after that is acceptance; if you would rather not, stop using it and ask us to close the account.

/TOS_14Governing law, and how to reach us

Governing law, and how to reach us

This agreement is governed by the law of the State of New York, United States, without regard to its conflict-of-laws rules, and a dispute arising out of it is heard by the courts located in New York, United States. Both parties submit to that jurisdiction.

Nothing in this section is intended to deprive a consumer of the protection of the mandatory law of their own country of residence, or of the right to bring a claim in the courts of that country where that law gives it to them.

Contact for anything about these terms: legal@superprotocol.com. For data questions, privacy@superprotocol.com. For security, security@superprotocol.com.