One line of code, and you run in the European Union.

Quickstart: your first API call

Create a key in your dashboard under Settings, API keys. You see the secret exactly once. After that we keep only a hash and the prefix. Put it in an environment variable, never in your repository.

curl https://api.europeanailab.eu/v1/chat/completions \
  -H "Authorization: Bearer $UNION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "eu-standard",
    "messages": [{"role": "user", "content": "Explain VAT reverse charge."}]
  }'

import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["UNION_API_KEY"],
    base_url="https://api.europeanailab.eu/v1",   # the only line that changes
)

resp = client.chat.completions.create(
    model="eu-standard",
    messages=[{"role": "user", "content": "Explain VAT reverse charge."}],
)
print(resp.choices[0].message.content)

Model tiers and context windows

You pass a tier in the model field, not a model name. That keeps your integration working when we replace the model underneath, and we will, because open-weight models get retired and overtaken.

Exceed a tier's context window and you get a 400 with the maximum in it: we never silently truncate your prompt.

Streaming responses

Streaming works the way you expect, with one important difference you need to know about because it concerns your money.

stream = client.chat.completions.create(
    model="eu-standard",
    messages=[{"role": "user", "content": "Write a summary."}],
    stream=True,
    # Without this there is no usage object and you cannot see the cost.
    stream_options={"include_usage": True},
)

for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")
    if chunk.usage:                      # final chunk
        print("\n", chunk.usage)

Cut a stream off halfway and you only pay for what was generated up to that point. If the broken connection means no usage chunk arrives, we estimate the consumption and correct it in the daily reconciliation with our provider. Such a line sits in your usage overview as estimated until it is settled.

Function calling and tool use

Tools follow the familiar schema. If a model produces a tool call that does not match your schema, we log it: in version 1 we do not silently correct it, because quietly repairing your output is worse than a visible failure.

tools = [{
    "type": "function",
    "function": {
        "name": "find_invoice",
        "description": "Look up an invoice by number.",
        "parameters": {
            "type": "object",
            "properties": {"number": {"type": "string"}},
            "required": ["number"],
        },
    },
}]

resp = client.chat.completions.create(
    model="eu-standard", messages=messages, tools=tools
)

Structured output with a JSON schema

Pass a JSON schema and validate the answer on your side. Our advice: always validate, even when the model usually behaves.

resp = client.chat.completions.create(
    model="eu-standard",
    messages=[{"role": "user", "content": "Extract the name and amount."}],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "invoice",
            "schema": {
                "type": "object",
                "properties": {
                    "name": {"type": "string"},
                    "amount_eur": {"type": "number"},
                },
                "required": ["name", "amount_eur"],
            },
        },
    },
)

Response headers: cost, tier and routing

Every response tells you what happened. This is not a debugging extra, it is the point: you can check for yourself that you got what you paid for.

Rate limits and spend ceilings

There is a limit per key and per account, plus the ceilings you set yourself. The latter matter more than they sound: without a ceiling, a leaked key or an infinite loop in your own code can drain a balance overnight.

• Requests per minute, per key and per account.
• Daily and monthly ceiling in euros, configurable per key.
• New accounts: a maximum of €25 per day for the first 48 hours, whatever the balance.
• If we see more than ten times your weekly average within an hour, we pause the key automatically and email you. Better one false alarm than an empty wallet.
• A request only starts when your balance is at least €0.05.

Error codes and what to do

Failed requests cost you nothing: if no usable answer came back, nothing comes off your balance.

Fair use and usage policy

We refuse instructions for weapons, malware and fraud, and material that harms children. Those refusals are final: we never forward such a request to another model to get an answer out of it anyway.

A model refusing a legitimate question about politics, history or a socially sensitive subject is a different matter: we treat that as a defect and send it to a model without that blind spot. You see it in the header X-EU-Routed: censorship.

One of our models is licensed with its own usage policy. That policy applies in full: rerouting does not get around licence terms.

Not sure whether your use case fits? Email hello@europeanailab.eu. An answer up front is cheaper for both of us than a blocked key afterwards.

Frequently asked questions

The questions that arrive by email before anyone opens a key: what the API is compatible with, what a request costs, and where it runs.

Moving an existing integration across is three lines: the key, the base URL and a tier name. The whole change, plus what does not port, is in switching in three lines. Still open? The rates per tier are on the pricing page, the model and licence behind each tier on AI transparency, and what happens during an outage on the status page. Two longer reads: what an AI API costs per token and where your prompts actually go.