OE OpenExpertsAI

Security

How we keep your data under your control

Each organisation gets its own private AI workspace. This page explains where it runs, what can leave the server, who can reach it, and how your data is deleted.

In short

Where your workspace runs

After a pilot, you choose one of two setups:

During the pilot, your workspace runs on a server that OpenExpertsAI operates, which we call the pilot server. It is not in your office, and it can host other organisations’ pilots too. Your pilot runs in its own separate workspace, as described below.

How each workspace is kept separate

Each organisation gets its own workspace: its own application instance, its own database, its own document libraries and its own user accounts. Clients’ documents are not mixed. This applies on the pilot server too.

Where the AI models run

The AI models are open-weight models, for example from the Qwen family, that run on the server itself. Your documents, chats and recordings are processed on that server. The two features that can send information out are described below.

We do not use your data to train AI models. The models on the server do not learn from your documents or chats.

What can leave the server, and who decides

Two features can send information off the server. Both are switches, and which ones are on is agreed with your organisation during setup. Separately, how people reach a hosted workspace, including through Cloudflare, is described under Who can access it.

Web search

Web search is an administrator switch. When it is on, search queries made from a user’s question are sent to a web search service to fetch public pages. Your documents are not sent, but a query can contain words from the question, such as a name.

Outside AI

Outside AI means a stronger cloud model from an outside provider. It is optional, off unless your organisation chooses it, and limited to the accounts your administrator allows.

Before a message is sent out, personal details such as names, ID numbers, phone numbers, bank numbers and addresses are automatically replaced with placeholders. What goes out is the conversation, including any document passages it contains, after this masking. The real values are put back into the answer on the server.

The masking looks for personal details, not unpublished figures or results. It catches common forms of personal details, but it is not perfect. That is why we agree staff rules with your organisation for what may be sent to outside AI.

Who can access it

People stay in charge

Backups, retention and deletion

We agree backup, retention and deletion arrangements with each organisation in writing.

If your organisation stops after the pilot, we delete its workspace and data and confirm this in writing.

Questions or a security concern

If you have a question about security, or think you have found a security problem, email [email protected].

This page is about client workspaces. For how this website handles the details you send us, see the privacy policy linked below.

See it work on your own data

Book a 30-minute demo. Bring a document your team works with.

Book a demo