SOKKAN Inference API — Terms & Data Policy

Version 2026-10-03. This page governs https://api.sokkan.ch, the OpenAI-compatible inference API that NINABOT Sàrl (Geneva, Switzerland, CHE-287.575.522) offers to resellers and routers (for example OpenRouter or Hugging Face Inference Providers). For the SOKKAN Cloud product, see the legal notice, the privacy policy and the trust page.

1. Data policy — zero retention

2. Where inference runs

Every model listed by GET /v1/models declares its datacenter country. Models marked CH run in Switzerland, on hardware owned by NINABOT in Geneva (region: geneva) or on GPU instances rented from Exoscale in Zurich (region: zurich), with no overflow elsewhere: if capacity is reached, the API answers 429 immediately instead of rerouting. Sub-processors are listed in the DPA. A model hosted in another country is declared as such in /v1/models before it receives traffic.

Traffic reaches us through Cloudflare (TLS termination and network protection); Cloudflare does not store request bodies.

3. Service

4. Acceptable use

Resellers remain responsible for their end users and must pass on usage rules at least as strict as the acceptable use policies of the model authors. Illegal content, attacks on third parties and attempts to disrupt the service are forbidden; we may suspend a key that causes them.

5. Data processing agreement

Personal data in requests is processed under the Data Processing Agreement (GDPR art. 28, Swiss FADP art. 9), which applies automatically and can be signed as a counterpart on request.

6. Liability and law

Model outputs are generated automatically and may be wrong; they are provided without warranty. NINABOT's liability is limited to the amounts invoiced for the last three months, to the extent permitted by law. Swiss law applies; place of jurisdiction: Geneva.

7. Contact

NINABOT Sàrl, Chemin de Riantbosson 19, 1217 Meyrin (GE), Switzerland — [email protected].