The same API calls, a different address.
An OpenAI-compatible API on open models in the BIT data centre in Ede, with the arrangements on your data in your contract.
The OpenAI API is a good product: strong models, a quick start, pay per token. For a development team that runs customer data, files or source code through a model, what happens on the other side matters too. With us your client changes its base URL, key and model name; the model runs in Ede, and there is no American company in the chain.
No training by default. But 30 days of logs.
With the OpenAI API there is no training on your data by default. That is well arranged. Less well known: prompts and answers are kept for up to 30 days for abuse detection. Zero data retention exists, but only after approval by OpenAI. Processing in the EU exists too, but only for approved projects, on certain endpoints and with a surcharge on newer models. So the variant without logs and within the EU is there, but not by default and not for everyone.
And OpenAI remains an American company. The CLOUD Act applies regardless of region, and in 2025 a US court ordered OpenAI to temporarily preserve API traffic as well, for customers without zero data retention. With us the model runs in Ede, no storage of prompts is a contractual arrangement, and there is no American company in the chain. The price for that is a smaller model.
The OpenAI API next to an endpoint in Ede.
The same calls, a different address, different arrangements. And honestly: a different model tier.
Does your application need the strongest model, or do you lean on web search and agents? Then stay with OpenAI and apply for zero data retention and EU processing. If your workload is text over your own data and you want to be able to explain the chain, Ede is the simpler answer.
OpenAI is a trademark of OpenAI. Xyphen IT is not affiliated with or endorsed by OpenAI. Comparison based on public information as of 6 October 2026; check the current terms with the vendor.
An endpoint your team already knows.
Your developers work with the API they are used to. The difference is where the model is and what is in your contract.
No frontier models, no web search, no agents. What you do get is an engineer you can talk to about quantisation, context and latency.
This fits if you:
- Have a working integration on the OpenAI API and mostly do text, extraction or RAG over your own data
- Have applied for zero data retention or EU processing at OpenAI and ran into approval, a surcharge or a limited model choice
- Want to be able to explain to customers or regulators who is in the chain, without exceptions per endpoint
- Want a fixed model version, so you do not have to retest your prompts after a silent model swap
From intake to first call.
- IntakeWhich task, which data, how much traffic and which models you use now. That leads to a proposal: shared, dedicated or a combination, with the model we recommend. If it does not fit, we say so.
- Contract and addendumData processing agreement plus the addendum on training and storage. Only then do you connect anything that carries production data.
- Endpoint and migrationYou get your endpoint and key. You change base URL and model name, and we walk through the first calls with you. Prompts tuned to a large model usually need some adjusting.
- In useYou see your usage; we keep model and hardware up to date. A model swap only happens by agreement, in an announced maintenance window.
You can test on the shared capacity with test data before switching production traffic.
Frequently asked questions about an alternative to the OpenAI API
How much of my code do I need to change?
Usually the base URL, the key and the model name. Chat completions, embeddings and transcription work through the same paths as at OpenAI. Features specific to OpenAI, such as web search or the agents API, do not exist with us; you need to place those elsewhere or drop them.
Are the models good enough for my application?
For classification, extraction, summarising, question answering with RAG and transcription, models of 24 or 27 billion parameters do well in practice. For open-ended reasoning about arbitrary topics and long, complex documents, a frontier model remains stronger. In the intake we test your task with you on test data.
What is the difference with zero data retention at OpenAI?
Zero data retention at OpenAI is an exception you apply for and OpenAI has to approve; without that approval, prompts are kept for up to 30 days. With us, no storage of prompts is the default, laid down in your contract. On top of that, OpenAI remains an American company to which the CLOUD Act applies; with us there is no American company in the chain.
Can I not simply arrange EU processing at OpenAI?
Sometimes. EU processing is available for approved projects, on certain endpoints and with a surcharge on newer models. If you qualify and you need the frontier model, that is a perfectly good route. With us Ede is the only region; there is nothing to apply for and nothing falls outside it.
What if you swap the model?
We do not do that without you. The model version is fixed and only changes by agreement, in a maintenance window we announce in advance. So your prompts and your tests keep holding until you decide to move to a newer version.
How quickly do I have a working endpoint?
Not within a few minutes; that is the honest difference with an API key at OpenAI. We start with an intake, then the contract, then the endpoint. How long that takes depends on the form you choose and whether servers are going in alongside.
Change your base URL and keep the rest?
Tell us what you run through the model, how much, and with which client. After an intake you receive a proposal with the model, the form and the arrangements on your data.