GUIDE / PRIVACY

Where is your data?
And who can access it?

Updated 3 October 2026

The short answer

Every AI assistant involves two places: where your data is stored (conversations, memory, files, keys) and where the thinking happens (the language model). Your own server decides the first: it is in your company’s name, and you can leave with everything. For the thinking, text does go to a language model; at the major providers this is processed (also) in the US. Who does that, what they keep and whether they train on it belongs in your data processing agreement.

Two places, per type of service

Off-the-shelf assistant (cloud)Assistant on your own server (Oneeno)
Conversations, memory and filesin the provider’s cloud, where it chooseson a server in your company’s name, in Europe
What goes to the language modelwhatever the service needsthe pieces needed for each question
Where the thinking happensat the provider; often (also) in the USat Anthropic (text) and ElevenLabs (speech), (also) in the US
Who administers itthe provideryou, with the installer as administrator
Leaving with everythingexport as far as the service offers iteverything is already with you, in standard files
One breach, many clientspossible: all clients share one environmentno: each client has its own server

What your own server does settle

Your conversations, memory, files, the logins the assistant uses and the backups are on a server in your company’s name. There is no shared database with other clients, so a breach at one client does not affect the rest. And if you stop, everything is already with you.

For security, the major cloud services are very good too: they have large security teams. They do not offer that in combination with a server that is yours alone and that you can take with you.

What your own server does not settle

To think and speak, text or audio goes to a language or speech model. At Oneeno that is Anthropic for text and ElevenLabs for speech; by default both process (also) in the US. The team works with Claude and switches between the levels of Claude, up to the highest; Anthropic does not use business API data for training. A good set-up sends only what each question needs, keeps audio only until the report is ready, and records per supplier what happens.

So by default the storage stays in Europe and the thinking does not. If you want everything in Europe, we organise that, within our service.

Tailored: you choose where the thinking happens

With a standard service the provider decides where the thinking happens. Tailored, you can choose per type of work. For sensitive work a European model, such as Mistral, which stores in the EU by default. Or an open model on your own GPU server, so no text leaves at all; at a European host such a server costs roughly €230 to €2,300 a month (excluding VAT), and the best open models are about four months behind, according to Epoch AI. Or Claude through the European region of a cloud partner, at a small premium. And for speech, a business contract with storage in the EU.

At Oneeno the team works with Claude and switches between its levels up to the highest. A European model as a second supplier is coming soon. If you want everything in Europe, we organise that, within our service; the conversation is then about which route fits best and what it costs.

Who can access it

With your own server: you, and the installer as administrator, at your request. Your assistant only acts after your approval on your own signed-in device; it can read your project folders but not change them. How that is secured in eight layers is on our privacy and security page.

Frequently asked questions

Does my data stay in Europe if I use an AI assistant on a European server?

The storage does: conversations, memory and files are on that server. The thinking not always: the text sent to the language model for each question is processed (also) in the US by the major providers.

Who can access my data if my AI assistant runs on its own server?

You, and the administrator at your request. There is no shared database with other clients. The language-model provider only sees the pieces sent with each question.

Is an AI assistant on its own server safer than ChatGPT or Copilot?

For security, the major services are very good too. Your own server removes the risk that one breach affects all clients and puts everything in your hands; in return, administration has to be in order: updates, monitoring and approval before every action.

Want to see what is stored where?

In an introduction we show it layer by layer.

Sources

  1. Anthropic, Privacy Center: how long do we keep your data (commercial), updated 1 July 2026
  2. Anthropic, data residency (API), retrieved 3 October 2026
  3. ElevenLabs, privacy policy, updated 20 May 2026
  4. Mistral AI Help Center, data use and training, 31 August 2026
  5. Hetzner, GPU servers, prices from 15 June 2026
  6. Epoch AI, open-weight models are about four months behind, 29 May 2026
  7. Nudge Security, on the breach at Klue, 23 June 2026
  8. Oneeno, privacy and security: eight layers
Next question in the guide

An assistant that calls and listens in.