Where inference runs, what is stored and which contracts you get. For the strictest requirements, see the EU site.
Trust controls
EU requests run only on contracted operators inside the EU (API gateway in Helsinki, GPU inference in Finland), and inference data never leaves the EU.
EU TIER
EU-only processing
EU requests run only on contracted operators inside the EU. Today the API gateway runs on Hetzner in Helsinki and GPU inference on Verda in Finland; client traffic uses TLS 1.3 and gateway-to-GPU traffic a WireGuard tunnel. See the sub-processor list.
RETENTION
No inference payload logs
Prompts and outputs are never stored or used for training. Usage-history settings apply only to account records such as token counts.
ACCESS
Keys stay server-side
API keys are issued in the dashboard and are never embedded in static pages or examples.
PROCESSORS
Named sub-processors
The sub-processor page names every provider with its purpose and location. Sign-in (Google) and payments (Stripe) never receive your prompts.