>> Sovereign AI inference · OpenAI‑compatible · EU
EU‑sovereign AI inference, OpenAI‑compatible.
The frontier open models behind one base URL - inference in Germany, your prompts, your jurisdiction, never trained on. Coming soon.
- BSI C5
- ISO 27001
- GDPR
- Deutsche Telekom T Cloud Public
- Germany · eu‑de
Point my OpenAI SDK at fremai - what do I change?
Just the base URL. Same OpenAI SDK, same request shape - switch models by name, no rewrites, no bespoke client.
client = OpenAI(base_url="https://api.fremai.eu/v1", api_key=key) client.chat.completions.create(model="glm-5.2", …)routed · eu‑de · prompts never stored
Which models can I call?
A curated catalog of frontier open models - call any of them by name against the same endpoint.
Switch this thread to DeepSeek-V4.
Done - same thread, same API, just model="deepseek-v4". No client changes, no redeploy. Drop-in means drop-in.
Where does my prompt actually run - and do you keep it?
Inference runs on EU-sovereign infrastructure in Germany; the control plane is on Deutsche Telekom T Cloud - no US processor on the path. Prompts and completions are never stored or trained on; only the metering metadata needed to bill you.
EU · eu‑de · metadata‑onlyHow am I billed?
€30 per seat / month - every plan includes all models and features. fremai is prepaid: each seat bundles €10/month of usage, pooled across your org, and you top up beyond it - paying only for the tokens you use. Annual saves 15%. Pay by card; billed in EUR by fremverk, a European company.
Does it work with opencode, and can I lock it down for my org?
Natively - first-class login plus a models.dev catalog, zero config from opencode. For the org: model allow-lists, per-key rate / budget / concurrency limits, a tamper-evident audit trail, and bring-your-own IdP SSO.
$ opencode auth login fremai
Frontier models. European jurisdiction.
One OpenAI-compatible base URL, the frontier open models, hosted in Germany - your prompts never trained on. Coming soon.