Security and Data Retention.
Last updated November 10, 2026
Zero data retention
Every request is processed, answered and forgotten. Your inputs and the model's outputs are discarded as soon as the response is sent. We keep request metadata, such as timestamps and token counts, for billing and abuse prevention. We never keep the content.
GPU providers are held to the same rule
Models run on GPUs from several providers. Every one of them passes our security review before it serves a single request, and zero data retention is a condition of that review. A provider that cannot meet it does not serve your traffic.
If your policy requires US-only capacity, we can restrict your account to providers in the United States.
Nothing is used for training
Your data does not train, tune or evaluate any model, ours or anyone else's.
Compliance
Questions
Before you send a single request, ask us anything about data handling, retention or a specific provider, and we will answer in writing.