Model and Provider Terms

1. Whose rules apply

FlintBeam does not make the models it sells. Every request is served by a model provider, through an account we hold with it, and the provider's rules travel with your request: they apply to what you send and to what comes back. This page says whose rules those are, what the provider sees, what we do not allow, and what happens to an account that ignores it. It sits alongside the Terms of Service and forms part of them.

The catalogue names the vendor behind each model, and a model page says which provider serves it. The list is open: providers appear and disappear, a model can be served through more than one route, and which route answers a particular request can change without notice — when a route is at capacity, when a provider fails, or when an Economy version's daily volume is used up and the official version takes over. The policy of whoever served the request applies to it. Where two policies disagree, the stricter one applies.

2. The providers and their policies

Most models are served by one of the following. Each link goes to the provider's own rules on what may be sent, and to its own statement of what it does with the data it receives.

• OpenAI — usage policies, how your data is used
• Anthropic — usage policy, privacy policy
• Google, for Gemini through Vertex AI — prohibited use policy, data governance
• Microsoft, for models on Azure OpenAI — code of conduct, data and privacy
• Amazon, for models on Bedrock — acceptable use policy, data protection

Other providers named in the catalogue — model makers and hosting services that publish their own APIs — apply their own usage and data policies, published on their own sites. Before you send anything sensitive to a model, read the policy of the provider named for it. We link what we can, but the provider's own page is the one that counts.

3. What the provider sees

The provider receives your request in full — the messages, any images or files, tool definitions and parameters — and returns the answer through us. We do not store either; the Privacy Policy says exactly what we keep. What the provider does with the request afterwards is governed by its policy, not ours: the large providers keep it for a limited time to monitor for abuse and state that they do not train on API traffic, and they process it wherever they run, most often the United States. Using a model through us does not add any protection the provider does not give, and it does not remove any the provider does give.

4. What is not allowed

Using a model through us does not put you outside its provider's policy. Anything a provider's policy forbids is forbidden here, whether or not it is repeated below, and a refusal from a provider is not something we can overturn on your behalf. In addition, on every model and through every route, you must not use the service:

• For anything unlawful where you are, where we are, or where the provider is, or to help anyone else do something unlawful.
• To create, request or distribute sexual material involving minors, in any form, generated or not. There is no threshold and no context in which this is tolerated, and we report it.
• To plan, promote or assist violence, terrorism, self-harm, or the development or use of weapons capable of mass harm, including instructions for chemical, biological, radiological or nuclear weapons.
• To create or spread malware, exploits or tools for unauthorised access; to attack, probe, overload or disrupt any system — this service, a provider, or anyone else; or to test credentials that are not yours.
• For fraud, scams, phishing, or any scheme to take money or access that is not yours; for impersonating a person, an organisation or a service; or for deceiving people about who or what they are dealing with, including presenting a model as a human where that matters.
• To harass, threaten, defame or bully anyone, to generate hate or incitement against a person or a group, or to produce intimate imagery of a real person without their consent.
• To collect, infer or expose personal data about people without a lawful basis; to profile or surveil people; to make automated decisions with legal or similarly significant effect on a person without meaningful human review; or to evade a lawful requirement to tell people they are dealing with a machine.
• To provide medical, legal, financial or safety-critical advice, or to operate anything where a wrong answer can hurt someone, without a competent person reviewing the output before it is acted on.
• To get around the limits: several accounts to escape a rate limit, a ration, a suspension or a price; funding an account with a payment method that is not yours; reversing a payment after spending what it bought; or any attempt to be served for less than a request costs.
• To extract the models themselves — weights, systematic reproduction of their behaviour — or to use what they produce to train or improve a competing model where the provider's terms forbid it.
• To send content you have no right to send: other people's confidential information, or material whose licence forbids it.
• To pass this access on to your own users, customers or systems without holding them to at least these rules and taking responsibility for what they do.

5. Your users

Your users are yours. If you resell, embed or otherwise pass on this access, you answer for what is sent through it as if you had sent it yourself, and you must hold your users to at least these rules. "A customer did it" is not a defence we can offer a provider, and it is not one you can offer us.

6. What happens to an account that breaks the rules

We suspend the account, and we do it at once rather than after a warning: by the time an attack or a fraud is visible, waiting costs someone else money. Suspension stops every key on the account immediately — the keys scoped to a plan along with the rest — and freezes whatever balance is left on it. Nothing is taken at that moment and nothing is paid out either; the freeze is what makes it possible to look at all. The one exception is a suspension caused by a reversed payment: there the balance is reduced by the amount reversed first, as the Terms of Service say, and may go below zero.

We may also refuse, limit or throttle individual requests or models without suspending the account, where that is enough to stop the problem, and we may do so without explaining which rule was engaged where explaining it would help someone get around it.

Then we review what happened, and the review ends in one of two ways:

• If we find the service was used unlawfully, to defraud us or anyone else, or in a way that put us in breach of a provider's rules — which includes funding the account with a payment method that was not yours and reversing a payment after spending what it bought — the account is terminated, and neither the remaining balance nor the unused part of any plan is returned. We keep the records needed to explain what happened, we may report it to the provider whose model was used and to the authorities, and we may refuse to open another account for the same person or organisation.
• If we find nothing of the kind, the account is unblocked. If you would rather leave anyway, the unspent balance is dealt with under the refund section of the Terms of Service.

We review without undue delay and tell you the outcome at the address on the account.

7. If you think we got it wrong

Write to support@flintbeam.com and say why. If we did, we say so and put it back. A review is reopened when there is something new to look at — a fact we did not have, an explanation that changes what the record means — and a repeated request without one is not.

8. Changes

This page can change, most often because a provider changes its own policy or because a new provider appears in the catalogue. The date at the top says when it last did, and the current version applies to every request from that date. These terms are governed by the laws of England and Wales, and the service is operated by CONNECT GLOBAL LIMITED, registered in England and Wales, number 16714288. Questions go to support@flintbeam.com.