Privacy Policy

Last updated: 15 November 2026

This English text is a convenience translation. The Indonesian version is the authoritative one and prevails if the two differ. Read the Indonesian version

This policy explains what data HematToken collects, what it is used for, and who it is given to. Operated by [ISI: nama badan usaha atau nama pemilik].

We have written it as specifically as we can. If there is a part you do not agree with, you should not use this service.

In short: HematToken does not store your prompts or the models' responses.

By default, none of the content you send and none of the responses models return is stored in our database — in any form, including fragments, summaries, or temporary copies.

What we store is a request log for monitoring, tracing, and billing: timestamp, model name, provider, status, token counts, cost, and latency. Metadata, not content.

There are two exceptions. First, the “request details” feature, which you switch on yourself for debugging: off by default, only you can read what it captures, and you can switch it off at any time. Second, a request REFUSED by our content-policy screening — the scanned request content is kept as evidence of a breach, readable only by our operators. Details of both are in section 1 below.

We also never use the content of your requests to train any model.

1. Request content and model responses

This is the question we are asked most, so it comes first. By default, HematToken does NOT store the prompts you send or the responses models return — not as text, fragments, summaries, or embeddings, and not as a temporary 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 from memory. No stage of that writes it to our database.

What we store for each request is a log of metadata only, for monitoring, tracing, billing, and fault diagnosis:

  • request time, provider, model name, status, and status code;
  • input and output token counts, estimated cost, latency;
  • which gateway key and which credential were used.

This log contains not one word of your request or response content. The same is true of our internal monitoring: the logs, metrics, and traces we ship there carry none of your request content.

The first exception is the "request details" setting on the Settings page — a debugging feature you control yourself:

  • it is OFF by default, and stays off if your account settings cannot be read for any technical reason;
  • while it is off, no request content is stored at all;
  • if you switch it on, a copy of the request and response is stored on your account — already filtered to remove values that look like credentials, and truncated if too long;
  • only you can read it, through your own account's dashboard;
  • what is stored is kept for at most 7 days and capped at the most recent 1,000 requests and 50 MB per account; anything past those limits is deleted automatically every hour;
  • you can switch it off at any time, and doing so stops any further capture immediately.

Switch it on only while you are actually chasing a problem, and consider first whether the data you send may be stored. To delete what has already been captured before those limits remove it, contact us.

There is one further exception, and we name it here so it does not come as a surprise: requests REFUSED by our content-policy screening. For those 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 — the part the screening actually read, roughly the first 16 KB of the request text, once values that look like an API key or a token have been removed. We keep it because an excerpt alone leaves nobody able to judge whether a breach really occurred, including you if you wish to dispute it. Where REVIEW BY AN AI MODEL is enabled, the portion of text around the match — at most about 2 KB, once credentials have been removed — is sent to a reviewer-model provider, which is a third party; and for a request judged not to breach, all we keep is the portion that was shown to that reviewer. It is readable only by our operators, is used for enforcing the Terms and meeting legal obligations, and is not used for anything else. The full terms are in “Content screening and evidence of a breach” in the Terms of Service.

Bear in mind that this is about what WE store. Your request content still has to reach the provider serving the model in order to be answered, and that provider's privacy policy applies there — see the section on who data is given to.

2. What data we collect

Account data:

  • your email address;
  • your password, stored as a hash — we never store the original and cannot read it;
  • registration time, email verification, and subscription status.

Provider credentials (BYOK):

  • the provider API key or OAuth token you connect, stored encrypted;
  • credentials are decrypted in memory only while a request is being processed, then discarded. A credential is never displayed in full again in the dashboard.

Your gateway keys:

  • stored only in a form that cannot be reversed to the original value, plus a few leading characters so you can recognise it;
  • the full value is shown exactly once, when it is created. If it is lost, we cannot recover it either — you need to create a new one.

Usage metadata, recorded for every request:

  • provider, model name, status, status code, timestamp;
  • input and output token counts, estimated cost, latency;
  • which gateway key and which credential were used.

Payment data is processed by a third-party payment provider. We receive the transaction status and the subscription period, not your card number.

3. Unlimited plan purchases

For purchases on the Unlimited site (without an account), we collect: the email address you type, the order number and status, the amount and payment reference from the payment provider, and the API key issued. The key is stored as a hash for authentication and in encrypted form so that it can be re-sent to the purchase email address; it is shown to you only on the receipt page and in the email.

For every request made with an Unlimited key, we record the same metadata as for any other request — time, model, status, token counts (cached tokens included), latency and fair-use tier — to count fair use, show your usage to you, and for monitoring. Request content and model responses are handled as set out in “Request content and model responses”.

The Unlimited site shows publicly the activity and usage ranking of each key under a random pseudonym (for example “ul-1a2b3c4d”) that does not contain, and cannot be turned back into, your email address, key or request content.

Your IP address is used temporarily for rate limiting and for limiting the number of real-time connections, to prevent spam and abuse; email addresses are used in hashed form to limit checkouts and key re-sends.

The only emails we send for Unlimited plans are: your key, renewal confirmations, a reminder before a period ends, and a key you asked us to re-send. Invoices are sent by the payment provider. Order and period records are kept for bookkeeping and dispute resolution.

4. What the data is used for

  • running the service: forwarding your requests to the provider you chose, to the provider behind a model we supply, or to whoever serves your Marketplace plan;
  • showing your usage, costs, and quota in the dashboard;
  • billing subscriptions, deducting balance for usage of models we supply, and applying the usage limits you set yourself;
  • keeping the service secure: detecting abuse, rate-limiting, and investigating incidents;
  • sending operational email — verification, password reset, and account notices.

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

5. Who the data is given to

AI model providers (BYOK models): every request you send through the gateway is forwarded to the provider you chose, along with its content. That data is then subject to that provider's privacy policy, not this one. We suggest you read it for each provider you connect.

AI model providers (models we supply): for models we serve ourselves, we forward your request to the provider behind the scenes using our credentials. The content still reaches that provider and is subject to their privacy policy. One model may be served by several providers with requests routed automatically, so the provider that receives any given request may vary. The list of providers for each model is shown on that model's page in the dashboard.

AI model providers (Marketplace plans): a request you send using a Marketplace plan is forwarded, with its content, to be processed. It may be processed by HematToken's own infrastructure or by another supplier serving the same model — that is, the partner that sells the plan, our own infrastructure, or another supplier offering the same model — and which of them processes it may differ from one request to the next. On the partner's or supplier's side their privacy policy applies; on ours, this policy applies, including our commitment not to store the content of your requests.

Content-policy reviewer model provider: where review by an AI model is enabled, the portion of request text around a pattern match — at most about 2 KB, once values that look like credentials have been removed — is sent to a model provider we appoint, to be judged. This happens only for requests that trip a screening pattern, not for every request.

Payment provider: to process subscription transactions and balance top-ups.

Email delivery provider: to send verification and password-reset email.

Analytics: the dashboard includes Google Analytics, which receives standard page-visit data including the address of the dashboard page you open. It does not receive the content of your requests or your credentials.

Infrastructure providers: where this service's database and servers run.

Beyond that, we only disclose data where required by law or by a lawful order from a competent authority.

6. Uncensored models

Using an uncensored model, as defined in the Terms & Conditions, does not change how we treat your data. We still do not store request content or model responses; we still record only the same metadata as for any other request; and the “request details” feature you switch on yourself applies in the same way. For a model whose content-policy screening we have switched off — which may include uncensored models — no request is refused by the screening, no part of a request is sent to the reviewer model, and no breach or review record is made; the second exception in “Request content and model responses” therefore does not arise for that model.

The content of a request to an uncensored model still reaches the provider serving that model, and that provider's privacy policy applies there. To the extent permitted by applicable law, we are not responsible for how that provider processes, stores, or uses the content of your requests.

We do not monitor, review, or screen the content of an uncensored model's answers. To the extent permitted by applicable law, we are not responsible in any form for personal data — yours or anyone else's — that you send to an uncensored model, that it produces, or that you distribute afterwards. Do not send other people's personal data without a lawful basis, and do not use uncensored models to expose, guess, or compile anyone's personal data. Use them wisely.

7. Cookies

We use cookies only as needed:

  • a session hint — it carries only a marker that you are probably signed in, used so pages do not flicker while loading. It contains no token and is not an authentication mechanism;
  • display and language preferences;
  • cookies set by Google Analytics.

Your own session token is stored in the browser's local storage, not in a cookie.

8. How long data is kept

Account data is kept for as long as your account is active.

Usage metadata is kept for your own history and billing.

Provider credentials are kept until you delete them from the dashboard. Deleting one takes effect immediately.

After an account is deleted, account data and credentials are deleted. Some records may be kept longer where needed for legal obligations, dispute resolution, or abuse prevention.

9. Your rights

You have the right to request a copy of your data, to request correction of inaccurate data, and to request deletion of your account and its data.

You can revoke provider credentials and gateway keys at any time directly from the dashboard, without contacting us.

Other requests can be sent to [ISI: alamat email resmi]. We may ask you to verify your identity before we act on a request concerning personal data.

10. Security

Provider credentials are stored encrypted. Passwords are stored as hashes. Gateway keys are stored in a form that cannot be reversed. Dashboard access requires a session that is re-validated on every request.

No system is completely secure. We cannot guarantee absolute security, and you are responsible for keeping your own password and gateway keys confidential.

If a security incident affects your data, we will notify you as required by applicable rules.

11. Children

This service is not intended for anyone under the age of majority under the law that applies to you, and we do not knowingly collect their data.

12. Changes to this policy

This policy may be updated. Material changes will be notified by email or by a notice in the dashboard, and the dashboard may ask you to accept the updated policy before you continue using it. The date of the last revision is shown at the top of the page.

13. Contact

Questions or requests concerning personal data can be sent to [ISI: alamat email resmi].