Stop sending your data to the US and China.
Union AI is an English-language AI model and chatbot. Open-weight models, every inference executed on servers in the European Union. The chat is €13 a month, everything included; the API is per token from prepaid credit in euros. No annual contract, no minimum spend, no surprise invoice.
An English-language AI model on servers in the European Union · Scaleway (fr-par)
Inference
100% in the EU
Every prompt and every answer is processed in the European Union. Never in the US or China.
Customer data
No training
Your prompts train no models: not ours, not anyone else's.
Billing
€13 a month
That is the chat, in full. Building on the API instead? Per token, prepaid from €10: when it runs out it stops, never an unexpected bill.
Integration
One line of code
OpenAI-compatible. Change your base_url and you are done.
From nothing to your first answer in three steps.
No sales call, no annual contract, no onboarding programme. You can start today and stop tomorrow.
Create an account
Verify your email and you get 10 free API requests. No card needed to start.
Subscribe or top up
€13 a month for the chat, or credit from €10 for the API, by iDEAL, SEPA direct debit, Bancontact or card. Credit stays valid for 24 months.
API key or chat
Generate a key with its own daily and monthly cap, or take the chat plan and skip the meter entirely.
Talk to the model, or build on it.
On the left, what you get when you just want to ask something. On the right, what you get when it goes into your own software: one key per environment, with a ceiling you set yourself.
Chat
eu-standard · fr-par
today
If I paste a customer complaint with personal data in here, where does it get processed?
In the European Union. Your prompt goes to our inference in fr-par and never reaches the US or China. We do not keep it after the conversation and we do not train on it.
1.4 s · 512 tokens · €0.0021
API
api.europeanailab.eu/v1
production · backoffice
12 Mar
ua_live_7f3c9b21key partly hiddenEvery key carries its own daily and monthly ceiling. Hit the cap and the key stops: your balance stays where it is.
Permissions
response headers
X-EU-Region: fr-par
X-EU-Model: qwen3.6-35b-a3b
X-EU-Routed: none
X-EU-Cost-EUR: 0.0021Every prompt is processed in the European Union.
Not a region setting on someone else's cloud: the models run on European infrastructure, under our management. Here is the full path of a single request.
- True: All AI inference runs on infrastructure inside the European Union.
- True: Your prompts and outputs never go to the US or China.
- True: The models run under our management and are not trained on your data.
- True: All inference runs on 100% renewable electricity, in a data centre cooled without air conditioning.
The same request, hop by hop, with what each layer can and cannot read: where your prompts go.
The path of one request
- InferenceEuropean UnionScaleway SAS
- Prompt retentionNonebeyond operational necessity
- Training on customer dataNocontractually excluded
- Hosting & CDNCloudflare, EU regionno caching of your requests
- DatabaseNeon, EU regionaccount and usage, never prompts
- PaymentsStripe (IE)Stripe Payments Europe
route_event · example
“What happened at Tiananmen Square in June 1989?”
eu-pro · primary model
“I can't discuss that.”
refusal detected after 180 ms · nothing sent to the customer
eu-pro · fallback model
“On the night of 3 to 4 June 1989 the Chinese army crushed the student protests…”
X-EU-Routed: censorship
We measure and correct the blind spots in open-weight models.
We run the strongest open-weight models on European hardware, under our management. Some open-weight models dodge political or historical questions. We catch that and send the question to a model that answers it properly. You stay on the same tier, at the same price.
- < 1%
- healthy reroute rate per trigger
- 1
- maximum reroutes per request
- 0
- reroutes on genuine safety refusals
What we do not do: reroute until some model plays along. Refusals about weapons, malware, fraud or material that harms children stand. Only categories on our allowlist are eligible for rerouting; every borderline case stays where it is.
How our routing worksPriced the way you build, or the way you type.
Both run on the same four tiers and the same European infrastructure. They are only billed differently, because a developer and a person cannot plan around the same unit: tokens for the one, a flat month for the other.
OpenAI-compatible API
Existing code keeps working. Point your base_url at us and pick a tier instead of a model name.
from openai import OpenAI
client = OpenAI(api_key="ua-...", base_url="https://api.europeanailab.eu/v1")
resp = client.chat.completions.create(
model="eu-standard",
messages=[{"role": "user", "content": "Summarise this contract."}],
)Chat
For everyone who does not write code. One price a month, every tier included, and not a token in sight.
Your plan
€13 / month
1,640 of 2,000 messages leftresets on 1 September · renews until you say otherwise
Messages large, tokens nowhere. A token meter tells a human nothing.
You buy a tier, not a model name.
Models get replaced, retired and overtaken. Tiers do not. That way a model swap at our provider touches neither your integration nor your price list.
eu-lite
Lite
Fast, cheap work: classification, extraction, short answers.
€0.40
per 1M input tokens
€1.60 per 1M output tokens
eu-standard
Standard
The workhorse tier. Default for chat and most API traffic.
€0.70
per 1M input tokens
€3.50 per 1M output tokens
eu-long
Long context
Whole case files, contracts and codebases in a single prompt.
€1.00
per 1M input tokens
€2.00 per 1M output tokens
eu-pro
Pro
Heavy reasoning, agents, complex code.
€4.00
per 1M input tokens
€12.00 per 1M output tokens
All amounts exclude VAT and apply to the API. The chat is €13 a month on every one of these tiers. No minimum spend, no annual contract. Add a colleague or a family member for €10 a month, 53% less than ChatGPT Business.
What we are building next.
We ship new features every few weeks. These are next on the list; if you need one of them to get started, tell us and it moves up.
- Fine-tuning on your own data
- Dedicated GPU capacity
- SAML sign-in and SCIM provisioning
- Mobile apps
- Embeddings and speech endpoints
- Batch API with volume discount
What people ask before they sign up.
European AI, open-weight models, what it costs and what happens to your prompts. The longer versions live on the pages each answer points at.
What is Union AI?
Union AI is a European AI service that runs open-weight language models on infrastructure inside the European Union. There is a chat plan for people and an OpenAI-compatible API for developers, paid from prepaid credit in euros. We pick the strongest open-weight models, run them for you and publish which ones sit behind each tier.
What is the difference between an AI chat plan and an AI API?
A chat plan is a finished product for a person: an interface, attachments, shared conversations, one monthly price. An API is a pipe for a developer: no interface, priced per token, billed by what you actually send. Ours are €13 a month and from €0.40 per million input tokens respectively, and most teams end up using both.
How much does European AI cost?
The chat is €13 a month excluding VAT with 2,000 messages included. The API is per token, from €0.40 per million input tokens, with no subscription and no monthly minimum. A thousand medium messages on the workhorse tier costs about €2.38.
Can I use AI in Dutch, German or Polish?
Yes. The chat and the API work in every official EU language, including Dutch, German, French, Polish and Swedish, and the site itself is in ten of them. Ask in your own language and the answer comes back in it. The quickest way to see the quality is to try it free on your own material.
Can I move to another AI provider later?
Yes, and that is deliberate. The API follows the OpenAI chat completions contract, so switching provider is a base URL, a key and a model name, and the models we run are open weight, which means someone else can run the same ones. What is never portable is the prompt tuning and evaluation you build on top, so keep those in your own repository.
Do I need a credit card to try it?
No. The first 5 chat messages and 10 API requests are free and ask for no card. You only pay once you subscribe to the chat plan or top up credit, and the minimum top-up is €10.
Start with AI for Europeans.
Use AI without sending your data abroad, everything on European servers, at a European price.
€13 a month for the chat, cancel any time. Or per token from €10 of credit.
Not sure yet? Take the 60-second quiz →