Cloud exit without chaos: what changes in 2027
From 12 January 2027, the EU Data Act bans defined switching charges for a cloud switch. The engineering work, however, remains. Here is how to prepare.
Jun 17, 2026
Sovereign AI services in Luxembourg: where SMOX fits
SMOX is a locally hosted AI platform operated by Datacenter Luxembourg. This guide explains what its API offers and how to assess a sovereign AI service.
Sovereign AI services in Luxembourg: where SMOX fits
A sovereign AI service is not defined by a label alone. It is defined by practical controls: where prompts and outputs are processed, who operates the service, how access is authenticated and what teams can verify before putting a workload into production.
SMOX is a locally hosted AI platform for European companies, operated by Datacenter Luxembourg S.A. Its current public offer centres on the Mixvoip Reasoning API, which lists Luxembourg as its processing region and provides chat, embedding and transcription capabilities through a private API key (1). For organisations handling sensitive internal information, that creates a concrete option to assess alongside public AI services. It does not, by itself, prove that a specific workload meets every contractual, security or regulatory requirement.
Sovereignty starts with questions that can be tested
Local processing matters when prompts may contain operational, customer or confidential business information. But location is only one part of the decision. A useful assessment should also establish who is responsible for the service, where its operational boundary sits and which records are available for review.
For an AI integration, ask four direct questions:
Where are requests processed? Confirm the region for prompts, generated responses, embeddings and audio files—not only the location of the company selling the service.
Who operates the platform? Check the contracting entity, support route and the parties that can administer the environment.
How does the application connect? A clear, documented API boundary makes it easier to keep credentials on the server and to separate application responsibilities.
What must be verified before release? Define the data classification, logging, retention, access and incident requirements for the workload rather than relying on a general product description.
These are engineering and procurement questions. They help a team make a controlled decision; they are not a substitute for legal or security review.
What the SMOX service offers today
The current SMOX service is presented as a private preview. Its developer documentation describes an OpenAI-compatible API at https://llm.mixvoip.ai/v1, authenticated with a Bearer API key over HTTPS.(2) The published model list covers a chat and reasoning model, a multilingual embedding model and a transcription model. The documentation also describes completed and streamed chat responses for the listed chat model.(2)
OpenAI compatibility can reduce integration work for a team that already uses a compatible client library. In the documented Python example, the application supplies the Mixvoip base URL and API key to the client, so requests are sent to the Mixvoip endpoint rather than the default OpenAI endpoint.(2) Compatibility should still be tested against the exact endpoint and model a team plans to use. It does not mean that every feature, model limit or operational behaviour of another provider is automatically available.
The most sensible starting point is a narrow proof of concept. Use representative but non-production data, list the models available to the supplied key, and measure response quality, latency and the integration path. Keep API keys in a server-side secret manager, not in browser code, repositories or logs, as the developer documentation advises.(2) Before a production rollout, confirm the current model limits, commercial terms and support arrangements with the service operator.(2)
The planned knowledge assistant
SMOX also describes a company knowledge assistant, but labels it as coming soon. The stated design is to connect sources such as emails, chats, recordings and documents; preserve provenance and access boundaries; and return source-aware answers through hybrid search.(1) That may be useful for organisations that need an internal knowledge layer rather than a model API alone. It should be treated as a stated product direction until its production scope, integrations and operating controls are documented.
For teams comparing options, SMOX is therefore best approached as two related tracks: an available API for chat, embeddings and transcription, and an announced knowledge-assistant capability. More information, developer guidance and private-trial access are available at SMOX.AI.(1)
Frequently asked questions
Does hosting AI in Luxembourg make a workload sovereign?
It is an important control, but not the whole answer. The organisation should also verify the operator, contract, access model, data flows and governance requirements for the specific workload.
Can an existing OpenAI-compatible application connect to SMOX?
The published documentation provides an OpenAI-compatible base URL and an example using the OpenAI Python client. Test the exact API calls, model IDs and limits that the application requires before deployment.(2)
Is the SMOX knowledge assistant available now?
The SMOX site currently marks the knowledge assistant as coming soon. The Mixvoip Reasoning API is presented separately as a private preview.(1)
