Sub-processors

Who receives your prompt

When you make a request, the text goes to whichever provider serves the model you reached: either a model provider's own API or a routing service that fronts several of them. That party processes your prompt in order to return a completion, under its own terms and privacy policy. This is unavoidable for any gateway: a request cannot reach a model without reaching whoever hosts it.

The specific providers in the routing pool change as supply changes — that is the point of the router, and a fixed list published today would be wrong next month. We will name the current set on request: privacy@freemodel.online.

Infrastructure

Accounts, payments and abuse prevention

Sign-in through GitHub or Google is optional. An account with an email and password uses none of those three, only the bot challenge.

What we do not run

Worth listing, because the absence is the point: there is no advertising network, no analytics platform, no session-recording tool, no error-tracking SDK that ships page content, and no data broker on either the site or the gateway. Nothing on these pages loads a script from a third-party domain except the bot challenge on the sign-in form.

Payments

The free tier needs no payment method and no card. Paid plans are bought on the account portal at alli.website, and the charge is collected by Creem, a third-party merchant-of-record provider. Creem is the seller of record for the transaction and takes the payment details at its own checkout; card data never reaches our servers, and we hold none of it.

Because Creem is the seller of record, its name is what appears on a card statement, and its receipt is where the Customer Portal link comes from. Cancelling and refunds both run through it: the portal cancels the subscription — which is what stops the next charge — and carries the support route for a refund or a charge that looks wrong. Access runs to the end of the period already paid for, and an approved refund goes back to the original payment method.

Changes

When a sub-processor is added or replaced, this page is updated with the date below. Because the routing pool turns over as supply changes, the provider list in the first section is described by category rather than frozen; ask us for the current set at any time.

The path a request takes

Which party sees what is decided by where the request goes, not by any setting you control. Your client sends the call; the gateway terminates it at its host, decides which candidate should serve it, and forwards the prompt to that provider or routing service. The answer returns along the same path. Only the forward carries the prompt itself; the hosting layer stores metadata at rest, and the web layer in front of these pages serves the documentation rather than the API.

• The provider or routing service that answers receives your prompt and returns the completion. No gateway can avoid this, and it is the party whose own terms govern what happens to the text while it is being processed.
• The gateway host sees the request as it is routed and stores the metadata afterwards: model, token counts, HTTP status, duration.
• Cloudflare handles DNS and reverse proxy for this website, with TLS termination for the pages. Its other appearance is the bot challenge on the sign-in form, which sees an IP address and browser signals.
• Alibaba Cloud DirectMail receives your email address if a transactional message is sent to it, and nothing beyond that.

The recipient list for a prompt is not fixed, and that is a property of the product rather than an omission. Six of the 199 listed models (3%) are modality routes that resolve to a provider at request time, and the pool behind them turns over as supply changes. Of the remaining entries, 77% are chat models and 9% generate images, spread across model providers and routing services. A list frozen today would be wrong next month, so the first section of this page is written by category and the current set is available on request.

Sub-processor FAQ