EUrouter Integration Guide
You are helping a developer integrate EUrouter into their application. EUrouter is a drop-in replacement for the OpenAI API, purpose-built for EU customers requiring GDPR compliance. It routes AI requests to 10 providers through a single OpenAI-compatible API at https://api.eurouter.ai .
Supported providers: OpenAI, Anthropic, Mistral, Scaleway (FR), GreenPT, Microsoft Foundry, Nebius, IONOS (DE), OVHcloud (FR), AWS Bedrock.
When helping developers, always load the relevant reference files from the references/ directory in this skill folder based on what they need. Do not load all references at once.
Authentication
EUrouter uses API keys in the format eur_<publicId>.<secret> .
Send the key via either header:
-
Authorization: Bearer eur_...
-
x-api-key: eur_...
For organization-scoped requests, add: x-org-id: <uuid>
Keys are created via POST /api/v1/keys and the full key is shown only once at creation. It cannot be retrieved later.
Drop-in Migration from OpenAI
EUrouter is designed as a drop-in replacement. Only two things change:
Python (openai SDK):
from openai import OpenAI
Before (OpenAI direct)
client = OpenAI(api_key="sk-...")
After (EUrouter) — change base_url and api_key
client = OpenAI( base_url="https://api.eurouter.ai/api/v1", api_key="eur_YOUR_KEY_HERE" )
response = client.chat.completions.create( model="gpt-4o", messages=[{"role": "user", "content": "Hello!"}] )
Node.js (openai SDK):
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://api.eurouter.ai/api/v1", apiKey: "eur_YOUR_KEY_HERE", });
const response = await client.chat.completions.create({ model: "gpt-4o", messages: [{ role: "user", content: "Hello!" }], });
Important: All existing OpenAI model names work as-is (e.g., gpt-4o , claude-3-5-sonnet , mistral-large-3 ). No model name changes needed when migrating.
Core Endpoints
All endpoints are prefixed with /api . Base URL: https://api.eurouter.ai
Endpoint Method Auth Description
/api/v1/chat/completions
POST Yes Chat completions (streaming, tools, vision, audio, reasoning)
/api/v1/completions
POST Yes Legacy text completions
/api/v1/embeddings
POST Yes Text embeddings
/api/v1/responses
POST Yes OpenAI Responses API compatible
/api/v1/models
GET No List/filter model catalog
/api/v1/providers
GET No List available providers
/api/v1/credits
GET Yes Check credit balance (EUR)
/api/v1/keys
CRUD Yes Manage API keys
/api/v1/routing-rules
CRUD Yes Manage named routing configs
/health
GET No Health check
For full endpoint schemas, load references/endpoints.md .
EU/GDPR Compliance
This is EUrouter's key differentiator. Add a provider object to any generation request to enforce compliance:
{ "model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}], "provider": { "data_residency": "eu", "eu_owned": true, "data_collection": "deny", "max_retention_days": 0 } }
Field Type Description
data_residency
string
Region code — "eu" , "de" , "fr" , etc. Only endpoints in that region
eu_owned
boolean
Only EU-headquartered providers (Scaleway, IONOS, OVHcloud, GreenPT, Mistral)
data_collection
"allow" | "deny"
"deny" excludes providers that use data for training
max_retention_days
number
Max data retention. 0 = zero-data-retention only
Ready-to-use compliance profiles:
// Strict GDPR: EU owned, no training data, zero retention, EU hosted { "eu_owned": true, "data_collection": "deny", "max_retention_days": 0, "data_residency": "eu" }
// EU hosted, cheapest price { "data_residency": "eu", "sort": "price" }
// Specific EU providers only { "only": ["scaleway", "ionos", "ovhcloud"] }
For all provider preference fields, load references/provider-preferences.md .
Model Fallback
Pass a models array to try multiple models in order. If all providers for the first model fail, EUrouter tries the next:
{ "models": ["claude-3-5-sonnet", "gpt-4o", "mistral-large-3"], "messages": [{"role": "user", "content": "Hello"}], "provider": { "data_residency": "eu" } }
Note: Model fallback is NOT supported for embeddings (vector dimensions differ between models). Only provider fallback (same model, different provider endpoint) works for embeddings.
Routing Rules
Create reusable routing configurations to avoid repeating provider preferences on every request:
// Create a rule once: // POST /api/v1/routing-rules { "name": "gdpr-strict", "provider": { "eu_owned": true, "data_collection": "deny", "max_retention_days": 0, "data_residency": "eu" } }
// Then reference it by name: { "model": "gpt-4o", "rule_name": "gdpr-strict", "messages": [{"role": "user", "content": "Hello"}] }
You can also reference by UUID with rule_id . Per-request provider preferences merge with the rule — request-level values take precedence.
Dry-run testing: Use POST /api/v1/routing-rules/dry-run to preview which providers satisfy a routing config without making a real request.
For full routing rules API, load references/routing-rules.md .
Model Discovery
GET /api/v1/models GET /api/v1/models?provider=scaleway GET /api/v1/models?supported_parameters=tools,vision GET /api/v1/models?category=embedding
Prices are returned as strings in the model's pricing object with a currency field. Credits are in EUR.
Response Headers
Every generation endpoint returns routing transparency headers:
Header Description
x-provider-id
Internal provider UUID
x-provider-slug
Provider name (e.g., openai , scaleway )
x-routing-strategy
Strategy used for routing
x-model-used
Actual model slug used
x-model-variant
Variant suffix if applicable (:free , :nitro )
x-fallback-count
Number of provider fallbacks (only present if > 0)
x-model-fallback-count
Number of model fallbacks (only present if > 0)
Streaming
Set "stream": true for SSE streaming. Format is OpenAI-compatible:
data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"delta":{"content":"Hello"},"index":0,"finish_reason":null}]}
data: [DONE]
For usage stats in the final chunk, add "stream_options": { "include_usage": true } .
Error Format
All errors follow:
{ "error": { "code": "NOT_FOUND", "message": "Model not found" }, "requestId": "uuid" }
Key error codes: UNAUTHORIZED (401), FORBIDDEN (403), NOT_FOUND (404), BAD_REQUEST (400), INSUFFICIENT_QUOTA (402), SERVICE_UNAVAILABLE (503).
Reference Files
Load these based on the developer's specific need:
Need File
Full endpoint request/response schemas references/endpoints.md
All provider routing preference fields references/provider-preferences.md
Routing rules CRUD + dry-run API references/routing-rules.md
Python code examples (openai SDK) references/code-examples-python.md
Node.js/TypeScript code examples references/code-examples-node.md
curl examples references/code-examples-curl.md