open models, EU-hosted, no retention.
your app talks to one endpoint. ozeye picks the provider, retries on failure, and never routes outside the EU.
smart routing
automatic failover with one endpoint. always routes requests dynamically by cost, latency, or availability.
zero retention
no prompt or completion is ever stored. zero-retention rules are strictly enforced instantly at our API gateway.
EU-only infrastructure
your data never leaves European borders. requests are strictly limited to EU-hosted providers, with zero non-EU fallback.
per-model pricing.
these are the prices you pay on ozeye. no subscription fees, no hidden platform charge.
| model | input / 1Min | cached / 1Mcached | output / 1Mout |
|---|---|---|---|
| DeepSeek V4 Flash 0731deepseek-v4-flash-0731 | €0.12 | €0.03 | €0.27 |
| DeepSeek V4 Prodeepseek-v4-pro | €1.68 | €0.42 | €3.35 |
| GLM 5.2glm-5.2 | €1.15 | €0.25 | €4.02 |
| GLM 5.3glm-5.3 | €1.34 | €0.25 | €4.22 |
| GLM 5.3 Flashglm-5.3-flash | €0.19 | €0.05 | €0.48 |
| Kimi K3kimi-k3 | €2.87 | €0.72 | €14.37 |
| MiniMax M3minimax-m3 | €0.38 | €0.10 | €1.92 |
more than an API.
- private chataccess all models through a secure, built-in chat interface. no logs, no prompt persistence, and completely cookie-free for maximum privacy.
- hosted notebookswrite code and inspect responses in stateful, private python environments. perform direct validation and evaluation inside your secure browser session.
- configurable presetssave model, system prompt, temperature, and parameters as reusable presets. reload a full configuration in one click. switch tasks without retyping settings.
privacy by architecture.
your data flows strictly inside the European boundary. ozeye secures your path with validated zero data retention across all active endpoints.
frequently asked questions.
exclusively in the EU. requests route only through EU-hosted providers (Mistral, Scaleway, OVHcloud, Inceptron, TensorX) with no non-EU fallback.
no. zero data retention is enforced at the ozeye gateway and every provider we route to carries a zero-retention agreement: content is never written to logs or persistent storage, and is never used for training. some providers briefly hold cache state derived from a prompt so that repeated prefixes cost less; the subprocessors page explains exactly what that means.
we track metadata only: model, token counts, and cost. no message content is ever stored.
all listed on /models, with automatic failover across EU providers.
prepaid credits, pay per token. no subscription, no card required to start. top up when you need.
there is nothing to cancel. pay only for what you use. credits are prepaid, never expire, and non-refundable. stop anytime.
change your base URL and API key. the API is OpenAI-compatible, so existing code works. see /docs.
US providers route through global server networks with non-EU fallback, store prompts for safety or abuse review, and sit under US jurisdiction (CLOUD Act, FISA). ozeye routes EU-only with no non-EU fallback, enforces zero retention at the gateway, operates under EU GDPR jurisdiction, and bills prepaid per token with no subscription or seat minimums.