Legal

Data Policy

Last updated November 3, 2026

TokenFactory, operated by Synvia Corp, does not retain prompts or outputs. What you send to the language model API and what comes back is passed through, not stored, and is never used to train a model. This page says exactly what we do record, and who else touches your requests.

What we do not keep

Prompts: messages, system prompts, images, tool definitions and tool results.
Outputs: completions, reasoning text and tool calls, streamed or not.
Request content in logs. Our application logs carry no prompt or output text. When a provider returns an error we log its status code and the first few hundred characters of its error message, which describe the failure, not your request.

A request lives in the memory of our gateway for as long as it takes to forward it and stream the answer back, then it is gone.

What we do record

One metadata row per completed request, so we can bill you and show you your usage:

Your user and API key identifiers, the model name and the time.
The number of input, cached and output tokens, and the amount charged.

Our web server also keeps ordinary access logs (time, IP address, path, status code) for security and debugging. None of it contains request bodies.

Training

We do not train models on your data, and we do not sell it or use it for advertising.

Moderation

We do not read or filter request content. Your use of the API is subject to our Terms of Service.

Other products

This page is about the language model API. Products that store content by design, such as generated video you download later or storage volumes you create for GPU clusters, keep that content until you delete it. See the Privacy Policy for the full picture.

Questions

Write to contact@zeus.rocks.