Terms of Service

HematToken is a gateway and router for requests to AI model providers. We provide a single endpoint that accepts several widely used request formats (among them the OpenAI, Anthropic, and Google formats), translation between provider formats, fallback across models and across accounts, token compression, usage accounting, and guardrails.

The service can be used in three ways, and all three can run at the same time:

(a) BYOK — you use your own credentials (an API key or an OAuth session) with each provider. For this route we do NOT sell tokens, quota, or model access, and the commercial relationship for model usage remains between you and that provider.

(b) Models we supply — some models we serve ourselves using our own credentials, and you pay for them per token from your balance. For this route we act as the seller of model access to you, under the "Balance and per-token payment" section below.

(c) Partner plans (Marketplace) — you buy a ready-made plan belonging to a curated third-party provider, and we hand it to you as a single connection in your account. Its terms are in the "Marketplace plan purchases" section below.

The subscription fee is a fee for this gateway software. It is SEPARATE from your balance: subscribing does not give you balance, and topping up your balance does not extend your subscription.

We treat the content of your requests as yours. HematToken does not store the prompts you send or the responses models return — not as text, not as fragments, summaries, or embeddings, and not as any copy that survives once the request has been processed.

Request content only passes through: it is received, forwarded to the provider serving that model, the response is returned to you, and it is released. What we record for each request is metadata only — timestamp, provider, model name, status, token counts, estimated cost, and latency — which we need for monitoring, tracing, billing, and fault diagnosis. That metadata contains neither request nor response content.

The first exception is the "request details" feature: an optional debugging tool that is OFF by default. While it is off, no request content is stored at all. If you switch it on yourself, a filtered and truncated copy is stored on your account, readable only by you, capped at the most recent 1,000 requests, and you can switch it off at any time. The second is a request refused by our content-policy screening, described in its own section below.

We do not sell the content of your requests and we do not use it to train any model.

These terms govern what WE store. To be answered at all, your request content still has to reach the provider serving the model, where that provider's own privacy policy applies. For BYOK models you choose that provider; for models we supply and for Marketplace plans, the list of providers is shown in the dashboard.

Full details are in the Privacy Policy, which forms an inseparable part of these terms.

You must provide a valid email address and keep your password confidential. You are responsible for all activity that takes place through your account.

You must be old enough to enter into a binding agreement under the law that applies to you.

One account is intended for one person or one business entity. You may create several gateway keys for your own use.

We may close or restrict new account registration at any time — during a limited trial period, for instance, or while we are dealing with an incident. Closing registration does not affect existing accounts, and does not prevent you completing email verification for a link we already sent you.

We may also invite prospective users through an invitation link sent to a specific email address. Invitation links are personal, valid once, and expire. Sending a new invitation to the same address invalidates the previous link.

This section applies when you connect your own credentials (BYOK). You represent and warrant that you are entitled to use every credential you connect, and that using it through the gateway does not breach your agreement with the provider concerned.

Some providers we support use subscription or OAuth sessions that are officially NOT licensed for use through a proxy or router. Connecting one may result in your account being restricted, suspended, or permanently banned by that provider. We flag providers of this kind in the dashboard before you connect them.

That risk is entirely yours. We are not liable for restriction, suspension, banning, loss of quota, or any other loss arising on your account with any provider.

New accounts get a 3 (three) day trial with every feature unlocked, with no credit card required.

Once the trial ends, features that create or change data — adding a provider, creating a gateway key, creating a combo, and sending requests through the gateway — are locked until you subscribe. Data you already have remains readable.

Subscription prices are shown on the pricing page and at checkout. Payments are processed by a third-party payment provider; we do not store your card details.

A subscription runs for the period you choose and is active from the moment payment is confirmed. Payments received are not refundable for the period in progress, including if you stop using the service before it ends, unless applicable law provides otherwise.

We may change prices at any time. A price change does not apply retroactively to a period you have already paid for.

Your balance pays for the models we supply ourselves. It is NOT used for BYOK models — usage of a BYOK model is billed to you directly by your own provider.

Balance is topped up in Rupiah and recorded in US dollars. The exchange rate applied is the rate we set and show you BEFORE you pay; the Rupiah amount on that invoice is what binds. We do not take a margin on the exchange rate — our margin is in the per-token price.

A top-up is a minimum of USD 1 and a maximum of USD 1,000 per transaction, in whole dollars. An unpaid top-up invoice expires by itself and creates no obligation for you.

Balance does not expire and is not forfeited with the passage of time.

Balance has no cash value, cannot be converted back into money, cannot be transferred, and cannot be exchanged for anything other than model usage on this service. Please top up according to what you actually need.

Referral discounts and point redemptions apply ONLY to subscription payments. Top-ups are always charged in full, with no deduction of any kind.

The per-token price of each model is shown in the Models menu and may be changed by us at any time. A price change does not apply retroactively to requests that have already completed.

Before a request is sent to a provider, part of your balance is held as an estimated cost. If the request is cancelled or fails BEFORE it is sent, that hold is released in full and you are not charged. If the request has been sent, the tokens consumed are still charged even if you did not receive the response in full — because the provider has already charged us for them.

So that requests do not fail mid-flight, your balance may go negative up to a limit of USD 1. That difference is deducted from your next top-up. Below that limit, requests to models we supply are refused until the balance is topped up.

One model may be served by several providers behind the scenes, and your requests are routed among them automatically. What you pay is the displayed price of the model, not the price of whichever provider happened to serve it.

If we suspend or terminate your account for breach of these terms, your remaining balance cannot be cashed out.

The Marketplace sells ready-made plans belonging to curated third-party providers (“partners”). We are the seller of those plans to you; the model access behind them comes from the partner and is subject to their systems.

Once payment is confirmed, the plan is delivered as a single connection in your account and can be used through the gateway immediately, like any other connection. You do not need to register with the partner yourself.

Every plan has a VALIDITY in days and a QUOTA (a number of tokens, or a number of requests). Both are shown on the plan card before you pay.

Validity is counted from the moment the plan is ACTIVE — that is, from when your payment is confirmed — not from when you start using it. Delaying use does not extend validity.

Some plans have a quota that is restored periodically (daily, for example); the rest are one-shot quotas that are not restored. This is shown on the plan card.

A plan stops serving requests when its validity expires OR its quota is exhausted, whichever comes first. Unused quota is not refunded, cannot be cashed out, and does not carry over to a later plan.

Plans do NOT renew automatically. If you want to continue, you buy a new plan at the price applicable at that time.

Plan usage is counted according to the partner's rules, including any per-model markup they apply. The number of tokens deducted from your quota may therefore differ from the raw token count on the model's response. We show the usage we record in the dashboard.

Plan prices may change at any time. A price change does not apply retroactively: a plan you have already bought continues at THE PRICE YOU PAID and for THE VALIDITY that applied when you bought it.

What CAN change while a plan is running is its quota, its refill schedule, and its rate limits — including on a plan you have already bought. Those follow the partner, and we apply their changes so that our accounting keeps matching theirs; without that, the quota we show you would differ from the quota you actually have on the partner’s side.

We may mark a plan or a partner as temporarily unavailable — while the partner is having an incident, for instance. During that time the plan cannot be bought; plans you already own are unaffected.

Partners are third parties. An outage, a change to their model list, a policy change, or a partner discontinuing their service is outside our control, and we do not guarantee the availability of any particular model during your plan's validity.

Plan payments received are not refundable, unless applicable law provides otherwise. If a plan could never be delivered to you at all, we refund the payment for that plan.

Plan purchases are SEPARATE from subscriptions and from balance: buying a plan does not extend your subscription and does not add to your balance, and vice versa.

Referral credit is granted under the terms in force at the time of the transaction and may only be used to reduce the SUBSCRIPTION fee on this service — not to top up your balance.

A user who registers using someone else's referral code receives a one-off discount on their first subscription payment, at the percentage shown during registration and at checkout.

Referral credit has no cash value and cannot be cashed out, transferred, or exchanged for money.

We may cancel credit obtained through fake accounts, repeated registrations, or any other unreasonable means.

You must not use the service to:

• carry out activities that break the law in Indonesia or in your own jurisdiction;
• breach the terms of service of any model provider;
• resell, rent out, or provide access to this service to others as your own product without our written consent;
• attempt to break in, place unreasonable load on the service, scan for vulnerabilities, or disrupt its availability for other users;
• process data you have no right to, or data subject to a confidentiality obligation that forbids this kind of processing.

You must also not request, generate, or process through this service:

• sexual material involving minors, in any form whatsoever;
• planning a terrorist attack, building a bomb or explosive device, or recruiting or funding for a violent extremist group;
• producing or weaponising a chemical, biological, radiological, or nuclear agent;
• human trafficking, forced labour, or forced prostitution;
• planning to kill or seriously harm an identifiable person, including how to conceal it;
• explicit sexual material, and tools that undress or sexualise a real person's photograph without their consent;
• writing ransomware, credential stealers, rootkits, or botnets;
• phishing pages and kits, stolen-card tooling, and intercepting one-time passcodes;
• requests for a method, dose, or technique for suicide or self-harm.

These prohibitions apply to EVERY account and CANNOT be switched off by any setting. The reason is not only compliance: most models on this service are served using shared credentials — our own, a Marketplace partner's key ring, or a giveaway pool — and a provider that detects abuse suspends the ACCOUNT, not the request. One prohibited request from one user can take that model offline for every other user.

You are fully responsible for the content of the requests you send through the gateway, for the output you receive, and for how you use it.

To enforce the prohibitions above, HematToken checks the text of your request against a set of patterns before the request is forwarded to a provider. This check runs on every request, for every account, and is not a setting you can change. The list of categories is visible at any time on the Guardrails page in the dashboard.

When a request is detected as breaching, it is REFUSED before it reaches any provider, and you are not charged for it.

AN EXCEPTION TO OUR COMMITMENT ON STORING REQUEST CONTENT. For requests refused by this screening only, we keep a breach record containing the time, the category, the name of the pattern that matched, the model addressed, the excerpt of the text that triggered the match, AND THE SCANNED REQUEST CONTENT ITSELF. This record is the only circumstance in which we retain your request content outside the “request details” feature you control yourself.

What is kept is bounded in two ways, and we state them so you know exactly what they are: (1) only the part the screening actually EXAMINED, which is roughly the first 16 KB of the request text — the rest is never read and never stored; and (2) values that look like an API key, a token or any other credential are removed first, so they are not retained. This content is not sent to our internal notification channel and is not copied into the delivery log behind it; it lives on the breach record alone.

We keep it because without it an accusation of breach cannot be checked by anyone — including by you, should you wish to dispute it. An excerpt alone is not enough: a few words lifted out of their sentence can read as a breach when the sentence around them is not one, and a decision about an account should not rest on that fragment. The record is readable only by our operators, is used for enforcing these terms and for meeting legal obligations, and is not used for anything else — not to train models, not for analytics, and not disclosed to third parties except where the law requires it.

To be honest about its limits: this screening is a pattern match, not a judging model. It catches what is stated plainly and can miss what is phrased carefully. A request passing the screening does NOT mean it is permitted — the prohibitions above apply in full regardless.

If you believe a legitimate request was refused in error, contact us with the time of that request.

The service is provided "as is" and "as available", without warranty of any kind, express or implied.

We do not promise any particular level of availability (no SLA), do not warrant that the service will be uninterrupted or error-free, and do not guarantee any particular result from any model.

This service depends on third-party providers. Outages, price changes, API changes, quota restrictions, or a provider discontinuing their service are outside our control.

We may carry out maintenance, change, or discontinue parts of the service at any time.

To the extent permitted by applicable law, we are not liable for any indirect, incidental, special, or consequential loss of any kind — including lost profit, lost data, business interruption, or lost opportunity — arising out of or in connection with your use of the service.

We are specifically not liable for: charges arising on your account with any model provider; actions a provider takes against your account, including banning it; model output that is inaccurate, offensive, or harmful; or loss caused by your credentials leaking through negligence on your side.

To the extent permitted by law, our total liability for all claims arising from this service is limited to a maximum of the subscription fees you actually paid us in the 3 (three) months before the event giving rise to the claim.

You agree to release and indemnify us, including our officers and the people who work for us, against all claims, losses, and reasonable costs (including legal fees) arising from your breach of these terms, your infringement of a third party's rights, or the content of the data you process through the service.

We may suspend or terminate your access, in whole or in part, if you breach these terms, if your use endangers the security or availability of the service for other users, or if payment obligations are not met.

For breaches that are not urgent we will try to notify you first. For urgent or unlawful breaches, suspension may be immediate.

You may stop at any time by ceasing to use the service and requesting account deletion via the contact below.

We may update these terms. Material changes will be notified by email or by a notice in the dashboard before they take effect.

By continuing to use the service after a change takes effect, you are taken to have accepted the updated terms. The date of the last revision is shown at the top of this page.

These terms are governed by the law of the Republic of Indonesia. Any dispute will first be resolved by discussion; failing agreement, it will be resolved by the competent court in [ISI: kota yurisdiksi].

If any provision of this document is held to be invalid, the remaining provisions continue in full force.

Questions about these terms may be sent to [ISI: alamat email resmi].