EU Endpoint
https://eu.api.llmapi.ai serves the same OpenAI- and Anthropic-compatible API as https://api.llmapi.ai, but every model call it makes runs on provider capacity inside the EU. Use it when your requests must be processed by models hosted in the EU.
WARNING
The guarantee covers the upstream model call only — see What the guarantee
covers before relying on it for compliance.
Switching to the EU endpoint
Change only the base URL. Your API keys, request paths and model IDs stay the same:
curl -X POST https://eu.api.llmapi.ai/v1/chat/completions \
-H "Authorization: Bearer YOUR_LLM_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-4o-mini",
"messages": [{"role": "user", "content": "Bonjour !"}]
}'
With an SDK, set the base URL to https://eu.api.llmapi.ai/v1 (OpenAI SDKs) or https://eu.api.llmapi.ai (Anthropic SDKs).
The X-LLMAPI-Region header described in Multi-Region is ignored on the EU endpoint: it always serves the EU.
Supported endpoints
| Endpoint |
Notes |
POST /v1/chat/completions |
Streaming and non-streaming |
POST /v1/messages |
Anthropic Messages format |
POST /v1/responses |
Stateless only — see Responses state |
POST /v1/embeddings |
|
GET /v1/models |
Lists only the models available in the EU, at EU prices |
GET /v1/me, GET /v1/credits |
|
Every other route — including /v1/messages/count_tokens, the /zc and /nvm payment routes, audio, images, video, OCR, batches, files and realtime sessions — answers 404 endpoint_not_available_in_region. Use https://api.llmapi.ai for those.
Models
Only models that a provider verifiably serves inside the EU are available. The Models page marks them with an EU badge and lets you filter by Available in EU; a model's detail page lists its EU price per provider. Requesting any other model answers 404 model_not_available_in_region. Routing, fallback and auto only ever choose among EU-available providers.
Tools
Client-executed tools (function and custom tools) work as usual. Tools that run on a provider's or our servers — web search, code interpreter, file search, computer use, MCP servers, plugins and web_search_options — answer 404 tool_not_available_in_region.
Responses state
previous_response_id, item IDs and encrypted reasoning items work on the EU endpoint: a follow-up is routed to the EU account that produced the earlier response. Continue a conversation on the endpoint that started it — a reference from the other endpoint is not routed to the account that holds its state, so the provider usually answers that it was not found.
Background mode (background: true) is not available on the EU endpoint, because its result can only be read from GET /v1/responses/{response_id}; such requests answer 404 endpoint_not_available_in_region.
Projects and own provider keys
The EU endpoint never uses your own provider keys:
- Credits projects work unchanged.
- Hybrid projects are served from our EU capacity and billed as credits.
- API keys (own key) projects answer
404 own_key_project_not_available_in_region.
Prices
EU prices can differ from global prices. A request to the EU endpoint is billed at the EU price of the provider that served it, with your organization's discounts applied as usual. Responses report that cost in cost_usd_* and it is what you are charged. Request logs show the billed region.
Guardrails
Guardrail rules apply as on the global endpoint. Rules that use an LLM judge run that judge on EU capacity; when the judge model is not available in the EU, the request fails instead of skipping the rule.
Caching
Provider prompt caching works as usual. The gateway's response cache keeps EU entries separate from global ones, and streamed responses are not served from the gateway cache on the EU endpoint.
Restricting which hosts a key may call
Organizations, projects and API keys can each restrict the hosts they may be used on (Allowed request hosts in organization policies, project settings and API key settings). A request must be allowed by every level that is set; for example, an organization limited to the EU host can never be used on api.llmapi.ai. A refused request answers 403 region_not_allowed. Changes can take up to a minute to reach every gateway.
What the guarantee covers
- Covered: the upstream model inference call. It is only sent to provider resources we have verified as EU-resident, and redirects to other hosts are never followed.
- Azure OpenAI: deployments inside the Azure EU Data Boundary, which can include EFTA countries.
- AWS Bedrock: the AWS EU regions and the destination regions of the
eu. cross-region inference profiles.
- Not yet covered: our gateway itself, request logs, caches, queues and storage, and error reporting. Observability exports you configure and Anthropic profile provisioning are also outside the guarantee.
This endpoint is not a claim of EU-27-only processing or of compliance with any regulation on its own. Review the scope with your compliance team.
Error codes
| Status |
Code |
Meaning |
| 404 |
endpoint_not_available_in_region |
The route or Responses background mode is not served on the EU endpoint. |
| 404 |
model_not_available_in_region |
No provider serves the model in the EU. |
| 404 |
tool_not_available_in_region |
The request uses a server-executed tool. |
| 404 |
own_key_project_not_available_in_region |
The project uses your own provider keys. |
| 403 |
region_not_allowed |
An organization, project or key policy does not allow this host. |
| 502 |
upstream_region_mismatch |
A provider endpoint was outside the EU, so it was never called. |
| 502 |
upstream_redirect_not_allowed |
A provider answered with a redirect, which is never followed. |
| 421 |
misdirected_request |
The request reached a gateway that serves a different host. |
/v1/messages returns these errors in the Anthropic error format with the same code.