Terms of Service

These terms are between you ("you", "Customer") and OPENFILL LTD, a company registered in England and Wales (company number 17360288, registered office 128 City Road, London, EC1V 2NX, United Kingdom) ("OpenFill", "we"). They apply when you create an account or use the API.

The Data Policy and Privacy Policy form part of this agreement. The commitments in the Data Policy, including that we do not train on your content and do not retain it by default, are contractual terms. They are not marketing statements.

If you reach OpenFill through OpenRouter or another aggregator, your contract for that traffic is with them, and their terms govern their leg of the request. These terms govern our execution of the request.

You must be at least 18. If you sign up on behalf of an organisation, you confirm you are authorised to bind it, and these terms bind both of you.

• You are responsible for everything done with your API keys. Keep them secret; do not buy, sell, share or transfer them.
• You are responsible for your own end users, and must impose terms on them at least as protective as these.
• You confirm you are not located in, resident in, or organised under the laws of a comprehensively sanctioned jurisdiction, and are not on any restricted party list. Do not use VPNs or proxies to evade this.

OpenFill sells inference as principal. There is a single seller, every transaction is prepaid, and for every request you send us directly your contract is with us alone, whichever capacity executes it.

Your request is an offer. Submitting a request creates no contract of sale. A contract forms only when we admit the request and dispatch it for execution. Until then it may wait, expire, or be rejected.

Your price is a maximum, and you usually pay less. Every request to a market model carries a bid. We publish a per-model clearing level. The amount charged for an executed request is the lower of the published level and your bid, multiplied by the normalized token usage of that request. You never pay more than you bid.

How the published level is set. The level moves toward the price at which the demand for a model meets its capacity, read across the orders waiting and running in a cycle. It is smoothed over time rather than moving directly with each cycle, it is subject to an absolute ceiling, and admission is not purely price-ordered: we apply a per-account fairness cap so one customer's backlog cannot monopolise capacity, and we exclude requests an account could not currently pay for. The level may be zero, in which case execution is free. There is no price floor and we do not guarantee any price.

The price lock is off by default. Without it your request is a classic limit order: it rests at the price you set and pays the level it clears at. With it we bid a quote that we publish and latch for a short interval, so the most you will pay is known before you send and carries a margin over the level. Setting price_limit does not turn the lock on: it caps whatever the request bids. Send price_lock: true on a request, or switch it on for the account, to lock a price.

A quote is a firm price. A quote sets the price limit for requests submitted while it is valid. It does not promise that capacity exists or that a request will fill. It is not an option or a reservation. A price-locked request gets a short execute window from the moment you submit it; if it has not started in that window it is rejected and not charged.

Some models are sold at a fixed price. A fixed-price model has three published prices per 1M tokens, one each for cache-hit input, cache-miss input and output, listed at GET /v1/models and on the model's page. An executed request to such a model is charged its measured tokens of each kind at the prices published when you submitted it; a batch line is submitted with its batch. A price limit or a price lock sent with it has no effect. It waits for capacity until its deadline, and requests waiting for that model take turns across accounts. We may change a model's fixed prices; a change applies to requests submitted after it is published, and a request still waiting when it takes effect is not executed and is not charged.

When you are not charged. You are never charged for a request that does not produce a result. Specifically:

• 402: your balance cannot cover the request at the current level, or at the model's fixed prices. Add credits, or, on a market model, wait for the level to reach zero.
• 408: the deadline you set elapsed while the request was still queued. You set this deadline, subject to a floor and ceiling; we do not cancel resting requests because the level moved.
• 429: a price-locked request did not start within its execute window, a request to a fixed-price model did not start before its deadline, or you hit a rate or queue limit.
• A request to a fixed-price model that was still queued when that model's prices changed.
• Any request that fails on our side.

Capacity is not guaranteed. We make no promise that capacity will be available for any model at any time, that any request will execute, or that the level will stay within any range.

Where and how requests execute. We execute requests on computing capacity that we operate ourselves and on capacity that we obtain from third-party providers, in whichever combination we choose. We may add, replace or remove providers and change the mix at any time, subject to section 10 and to the notice the Data Policy gives for sub-processor changes. A provider that executes a request does so for us, under section 10 and the Data Policy. We do not publish which provider, machine or country executed a given request, and disclose it only where the law or our agreement with you requires. No request is contingent on a particular provider, machine or country unless we have agreed that with you in writing. For requests you send us directly, your contract is with us alone; this does not affect any right you have under the law or under a transfer safeguard.

Our records govern. Charges are computed from usage as measured by our systems, and our records are the authoritative record of usage and charges. We may correct metering, clearing and pricing errors after the fact, including after a charge has settled, by crediting or debiting your balance. We will tell you if we do.

We use the words "order", "bid", "market" and "clearing" because they describe the pricing mechanism accurately. They do not describe a financial product, and the following are the operative facts:

• OpenFill is the only seller. You are not trading with other customers, and we do not match buyers to sellers. Where a request executes on a provider's capacity, we buy that capacity as principal and sell inference to you; the provider sells nothing to you.
• Everything is prepaid. We extend no credit. Nothing is margined, leveraged, financed or settled on a deferred basis.
• There is no short side, no resale, no secondary market and no cash settlement. The only thing delivered is inference, and it is delivered when the request executes. Delivery is never deferred to a future date fixed in advance.
• Your balance and any admitted request are non-transferable and cannot be assigned or sold.

The parties agree these transactions are contracts for the sale and delivery of computing services for prompt delivery, with no deferred delivery date, and that none is intended to be a security, a commodity futures contract, a swap, a derivative, or any other financial instrument. OpenFill is not an exchange, a trading venue, or a broker.

The published level, history and charts are informational, derived from our own execution data, smoothed, and may be delayed. Do not rely on them as a benchmark or index for any purpose outside OpenFill.

Because the level responds to demand, the following are prohibited:

• Submitting, modifying or cancelling requests with the purpose of moving the published level rather than obtaining inference.
• Coordinated or multi-account bidding intended to raise or depress the level.
• Using multiple accounts to evade the per-account fairness cap, rate limits or queue limits.
• Acquiring credits or placing requests for speculative purposes rather than your own consumption.
• Reselling, sublicensing or brokering capacity, requests, quotes or credits, other than through an aggregator we have authorised (OpenRouter is authorised).

You pay in advance, beyond any credit we grant without payment. Your balance is an amount denominated in US dollars, redeemable only for OpenFill services. It is not a deposit, legal tender, a stored-value instrument or a payment instrument. It earns no interest. It is non-transferable and non-assignable, and may be used only by your account.

• Every new account is opened with $1 of credit as a trial, granted once per person and without payment. Credit we grant without payment is not refundable and cannot be paid out, and we may withdraw it from an account opened to obtain it more than once.
• For an account opened through another customer's referral link at sign-up, we grant $5 of referral credit without payment to each of the referring and referred accounts once the referred account has spent $25 on inference, once per referred account. The code must be applied at sign-up, within the first hour after the account is opened, and each account may have only one referrer. You may not use your own code. Referral credit is subject to the same restrictions on refunds and payouts as the trial credit above. We may withhold or reverse referral credit obtained through self-referral, duplicate accounts or accounts opened for the purpose of obtaining that credit.
• Credit is sold at face value in US dollars. What you pay is what is credited.
• Extra credit on a purchase, by the amount paid in one purchase: 1% from $25, 2% from $50, 3% from $100, 4% from $200, 5% from $400, 6% from $800, 7% from $1,600, 8% from $3,200, 9% from $6,400, 10% from $12,800. The highest tier reached applies.
• We may agree a higher percentage with an individual customer, which then applies in place of the tier, or extra credit on a single purchase. Extra credit is granted without payment, under the trial's refund and payout restrictions. We may withhold or reverse it from accounts opened to obtain it.
• A card processing fee of 5% of the amount is added to every purchase and shown before you pay. The processor keeps it, so it is not refunded.
• Charges are applied against your balance and can never take it below zero. You can never owe us money for inference.
• When a market model's level is zero, execution on that model is free and proceeds even with a zero balance. A positive balance is needed only to execute at a paid level or at a model's fixed prices.
• Before admitting a request we estimate its maximum cost and treat that amount as committed while deciding whether to admit further requests. This is an internal admission-control estimate. It is not a hold or authorisation on your funds, and it is released when the request finishes.
• If you enable auto-recharge, you authorise us and our payment processor to charge your saved payment method until you turn it off.
• Auto-recharge runs when your balance falls below your threshold. It buys the credit amount you set, or the shortfall to the threshold if larger.
• Extra credit and the card processing fee both apply to an auto-recharge as they do to a top-up of the same amount.

Refunds. You may request a refund of your unused balance by emailing [email protected]. We will refund the unused portion of what you have paid us to your original payment method, less any amount already consumed, and a refund never exceeds what you have paid.

A refund is of the unused portion of what you paid for credit. Extra credit and the card processing fee are not refundable.

We do not forfeit unused balances on account closure. Refunds are processed manually and require an active account, so request one before requesting deletion: once an account is deleted we can no longer verify your identity or reach your original payment method. If we terminate your account for breach of section 7, we may withhold a refund to the extent your breach caused us loss, and we will explain why.

You may not use OpenFill to:

• Generate child sexual abuse material. This is absolute. It covers text as well as images, AI-generated as well as real material, and attempts as well as successes. We will report it to the relevant authorities and preserve the evidence.
• Break the law, or help anyone else do so.
• Create or distribute malware, ransomware, or tools to compromise systems or critical infrastructure.
• Generate non-consensual intimate imagery, or sexual content involving minors.
• Harass, threaten, defame, or incite violence against anyone.
• Run disinformation, impersonation, or coordinated inauthentic behaviour campaigns, or generate political content designed to deceive about its origin.
• Infer or assign characteristics such as race, political opinions, religion, sexual orientation or health status about identifiable people, or perform biometric categorisation or emotion inference in workplaces or schools.
• Perform social scoring, predictive policing based on profiling, or untargeted scraping to build facial recognition databases.
• Manipulate people through subliminal or deceptive techniques, or exploit vulnerabilities of age, disability or social situation, so as to cause harm.
• Provide medical, legal or financial advice to third parties without qualified human review, or make consequential decisions about employment, credit, housing, insurance, education or benefits without meaningful human oversight.
• Submit special category data (health, biometric, racial or ethnic origin, political opinions, religious beliefs, sex life or sexual orientation) or criminal offence data without a lawful basis for doing so.
• Submit protected health information, payment card data, or government identifiers. Our systems are not configured for them and we offer no BAA.
• Circumvent rate limits, quotas, safety measures or geographic restrictions.

Enforcement. We may throttle, suspend or terminate access, and report to authorities where the law requires. Where practical we will warn first and act proportionately, but we may act immediately where the law requires it, where there is a risk to others, or where the breach is severe. Report abuse to [email protected].

Good-faith security and safety research is welcome. Contact us first, do not use real third-party data, and do not degrade the service for others.

As between you and us, you keep all rights in your inputs and own the outputs generated for you, and we assign to you whatever rights we may have in them. Two qualifications apply: outputs may not be unique, because similar prompts can produce similar results for other customers; and we do not warrant that an output does not infringe a third party's rights.

You grant us a limited licence, which we may extend to a provider that executes requests for us, to process your inputs and outputs solely to execute your request, provide the service to you, and comply with law. This licence does not extend to training, publication, or any other use.

You are responsible for evaluating outputs before relying on them. They can be wrong. Do not use them as a sole basis for decisions affecting health, safety, legal or financial matters.

You warrant that you have the right to submit your inputs, including any personal data in them, and that doing so does not breach any law or third-party right.

We serve open-weight models that we did not create. The models page lists the ones we serve at any time, and each model's page names its developer and links the published weights, where the licence that accompanies them is stated. Your use of the models and their outputs is additionally subject to that licence, including any use restrictions in it. Where that licence is stricter than these terms, it governs. We may change, add or withdraw models, and will give notice of removals where we reasonably can.

For the content of your requests we act as your processor and you are the controller. We process it only on your documented instructions, being your API calls and account settings. Our staff are bound by confidentiality. We use sub-processors, including the providers of the capacity on which requests execute. The Data Policy names those it names, describes the rest by role, and says what each can reach. You give us general authorisation to engage them. We remain responsible for them, bind each of them by written terms, and give notice of changes as the Data Policy sets out, so you can object.

We will help you respond to data subject requests and meet your security and breach obligations, taking into account what we actually hold, which by default is nothing. If we become aware of a breach affecting your content we will tell you without undue delay and in any event within 48 hours of becoming aware. On account closure we delete the content we hold within 30 days. Your content may be processed in any country in which we or our providers operate, and your API calls are your instruction to execute each request on the capacity section 3 describes, wherever it is located. Where personal data goes to a provider named in the Data Policy, we rely on adequacy regulations where they apply and otherwise on the Standard Contractual Clauses with the UK Addendum in that provider's agreement, and will provide a copy on request with commercial terms redacted. Content sent to a withheld provider to execute a request goes encrypted in transit, without any identifier of yours, under written terms that confine its use to executing the request and operating the service that does so; you are responsible for ensuring that the personal data in your content may be processed in that way.

On reasonable notice we will provide information needed to demonstrate compliance with this section. Where you need an on-site audit, it is at your cost and no more than once a year unless a regulator requires otherwise.

For US customers: we process personal information only to provide the service, and we do not sell or share it, or retain, use or disclose it for any purpose outside our direct business relationship with you.

The service is provided on an as-available basis. There is no uptime SLA and no service credit scheme. Live uptime measurements are published on the Models page. We may change, suspend or discontinue parts of the service; where a change is material and adverse we will give reasonable notice.

To the fullest extent the law allows, the service is provided "as is" and we disclaim all implied warranties, including merchantability, fitness for a particular purpose, non-infringement, accuracy, and uninterrupted or error-free operation. We do not warrant that outputs are accurate, complete, current, lawful, or suitable for your purpose.

Nothing in these terms limits liability that cannot lawfully be limited. That includes liability for death or personal injury caused by negligence, for fraud or fraudulent misrepresentation, and, if you deal as a consumer, your statutory rights.

Subject to that, neither party is liable for indirect, incidental, special, consequential or punitive damages, or for lost profits, revenue, data, or business, however caused.

Subject to that, each party's total aggregate liability arising out of or relating to this agreement is limited to the greater of (a) the amounts you paid us in the twelve months before the event giving rise to the claim, and (b) one hundred US dollars. Multiple claims do not expand this limit. These limits apply even if a remedy fails of its essential purpose. This paragraph does not limit your obligation to pay amounts due, or your indemnity below.

You will indemnify us against claims arising from your inputs, your use of outputs, your applications, and your breach of these terms or of the law.

You may close your account at any time. We may suspend or terminate access where the law requires it, where you breach these terms, or where necessary to protect the service or others. On termination your API keys stop working, queued requests are not executed, and you may request a refund of your unused balance under section 6.

We may change these terms. For material changes we will give at least 30 days' notice by email to account holders before they take effect, and you may close your account and take a refund of your unused balance if you do not accept them. Other changes take effect when posted. Continued use after a change takes effect is acceptance.

This agreement is governed by the law of England and Wales, and the courts of England and Wales have jurisdiction. If you deal as a consumer, you keep the benefit of any mandatory protections of the law where you live, and you may bring proceedings there.

If any provision is unenforceable, the rest stands. Our not enforcing a term is not a waiver of it. You may not assign this agreement without our consent; we may assign it to a successor to our business. There are no third-party beneficiaries.

OPENFILL LTD, 128 City Road, London, EC1V 2NX, United Kingdom. General enquiries [email protected], data protection [email protected].