Data retention

Zero data retention is on for every account and every request.

• Prompts and outputs are processed in memory and discarded when the response is sent.
• They are never written to disk or logs, and never used to train or evaluate models.
• We keep only request metadata for billing: key, model, timestamps, token counts, latency and cost.

Prompt caching keeps the model’s internal state for recent prompt prefixes (the KV cache) in GPU memory for up to 1 hour, scoped to your account. This is computed data, not your text, and it is never written to disk. See Prompt caching.

The Zero Data Retention Policy is the binding version. For account and billing data, see the Privacy Policy.