Why Frontière AI?
Frontière AI is the European gateway to open AI models — an EU sovereign AI gateway through one prepaid OpenAI-compatible API. This page lays out exactly what that means — and what it does not — so your legal and engineering teams can decide on facts.
Five pillars
What Frontière AI is built on
01
European jurisdiction
Know where every request is processed. The 18 sovereign models run at EU providers with no non-EU capital control — OVHcloud and Scaleway today. Dedicated EU servers of our own are available on request; no model is served from them yet. The operating company is French: the corporate structure is designed to keep sovereign workloads outside US jurisdiction.
02
One API
One key, one endpoint, dozens of models. The API is OpenAI-compatible: point base_url at Frontière AI, swap the key, and your existing SDK, scripts and agents keep working. Switching models is a one-field change, not a migration.
03
Per-model verification
Six criteria per model — license, provenance, security, GDPR assessment, benchmarks and API testing — each ticked only once we have run, recorded and passed the check. 11 of the 25 live models pass all six today. What exactly did we verify? The checklist is public, box by box, on the model pages.
04
No vendor lock-in
No proprietary SDK, no account per model vendor, no rewrite when you switch models. The API speaks the OpenAI protocol, so the rest of your stack — LangChain, n8n, Vercel AI SDK, Open WebUI, LibreChat — plugs in unchanged.
05
Transparent routing
EU sovereign vs fast access is disclosed on every model card and in every API response. When a frontier model is only reachable through US infrastructure, it is labeled “fast access”, never “sovereign” — you choose per call, with full information. A model labeled sovereign never leaves the sovereign lane by itself: any fallback stays with sovereign providers unless you explicitly allow the others, and the x-routed-provider header names the provider that served the call.
Built for European enterprises
Four workloads where jurisdiction actually matters
The homepage speaks to developers; this section is for the people who sign off: CTO, CIO, DPO, CISO, AI lead. Four recurring cases, and what changes when the infrastructure is EU sovereign.
AI agents
Run your existing agents on European infrastructure. n8n, LangChain, Vercel AI SDK, Open WebUI and LibreChat all work through the same OpenAI-compatible endpoint — no rewrite, no new SDK.
Internal copilots
Protect employee prompts and company documents. Frontière AI writes no prompt and no completion to its database, on any model — only the token counts billing requires. On the sovereign models OVHcloud states zero data retention and Scaleway keeps no prompts except requests that cause errors or abuse, up to two weeks; on fast-access models, provider-side retention follows the serving provider's own policy.
Customer-facing AI
Deploy AI without automatically routing company data through US providers. Sovereign models keep inference under EU jurisdiction; anything else is labeled before you call.
Regulated workloads
Make infrastructure and model jurisdiction explicit. Per-model jurisdiction labels, and the fact that we store no prompt at all, give your DPO a factual answer instead of a marketing promise.
How we compare
Frontière AI vs OpenAI, Anthropic and OpenRouter
The same comparison your procurement team would run. Where a cell depends on deployment choices, the footnote applies.
| Criterion | Frontière AI | OpenAI | Anthropic | OpenRouter |
|---|---|---|---|---|
| EU sovereign infrastructure | ✓ | ✗ * | ✗ * | ✗ * |
| Open-weight models | ✓ | Limited | Limited | ✓ |
| One API | ✓ | ✓ | ✓ | ✓ |
| Model jurisdiction visible | ✓ | Limited | Limited | Variable |
| Sovereignty verification | ✓ | ✗ | ✗ | ✗ |
| Prepaid, no subscription | ✓ | ✓ † | ✓ † | ✓ |
* The cross is about corporate control, not server location. These vendors document regional processing options that vary by plan and platform, and are worth checking on their own pages against your policy — but the operating company stays US-controlled and subject to US legal process, and that is exactly what the sovereign label is about. The Frontière AI column describes its 18 EU sovereign models; its fast-access models are served by US companies and labeled as such before you call them.
† OpenAI and Anthropic bill their APIs from prepaid credit too, with no subscription required; monthly invoicing exists on some accounts at both. Checked on their own billing pages on 2026-09-23.
Frontière Verified
What exactly did we verify?
Six criteria, and a score per model rather than a blanket promise: 11 of the 25 live models pass all six today. The figures below count the live models we have confirmed on each criterion; a model's page shows its own checklist, box by box, once we have started checking it. A checkmark means we confirmed it ourselves — a missing one means not confirmed: not checked yet, or checked without passing. The same labels are returned in the API response.
Run your first call under EU jurisdiction today
Create an account, top up €10, change two lines. One key for the whole catalog.