Security and data retention

Transport and access

• All public traffic runs over HTTPS; HTTP requests are permanently redirected to HTTPS.
• Every API request is validated against the key: expiry, allowed-model list, token limit, per-minute and per-hour request limits, concurrent stream count.
• Repeated attempts with invalid keys from one address temporarily block that address.

What the logs keep

For each request the service stores metadata: time, path, response code, model, latency and token counts. The most recent request and response bodies are additionally kept for diagnostics.

There is no long-term prompt archive: only the most recent requests per key are available and new records push out old ones. Size is capped at 32 KB for regular requests and up to 8 MB for image generation; headers at 8 KB.

Practical consequence: do not put anything in prompts that must not end up in a diagnostic log. The legal terms of data processing are in the privacy policy.

Handling keys

• A key is visible only in its owner's account.
• Documentation examples use the placeholder sk-your-tokenator-key — no real keys appear on the site.
• Do not keep a key in a repository or in a config that gets committed to git. Pass it through environment variables.
• If a key may have leaked, contact support and stop using it.

Isolation between customers

Logs and statistics are bound to a key: one key's requests are not visible to the owner of another. Internal upstream provider identifiers and their keys never reach client-facing data, and provider error texts are replaced with a generic message.

Server-side requests to external addresses that originate from user data go through address validation: internal and private addresses are blocked and redirects are re-validated.

FAQ

Read next