Private AI for lab work. Unpublished results stay in your lab's own workspace.

What your team can do

• Mask figures before a table goes out

  Before a results table goes to a partner, the assistant masks the unpublished figures. A person reviews each one, then exports the partner copy.

• Questions on SOPs, methods and reports

  Ask about your SOPs, method documents and reports. The assistant works from your lab's own library and cites the source, and the source panel shows the passage behind each answer.

• Check figures in unpublished data

  Ask about the figures in results you have not published yet. Each answer points to its source, so you can check it against your data before you use it.

• Notes from lab meetings

  Upload a recording of a lab meeting. It is transcribed on the server, then drafted into notes for a person to edit and confirm.

• Draft reports from your results

  Draft report text from your own results without sending the data to a public AI. A researcher edits the draft and checks every figure.

• A channel for each study

  Colleagues discuss a study in a team channel and bring an assistant into the thread by typing @ and its name.

An example

A researcher needs to send a results table to a partner, but three of the rates are not published yet. They ask the figure masking assistant to prepare a partner copy of batch12_results.xlsx. The assistant masks the three unpublished rates and keeps the measure names and methods, so the partner can see what was measured but not the results. A person reviews each masked figure and approves the copy before it is sent.

How your data is protected

• Your lab's own workspace. Your lab gets its own workspace, with its own application instance, database, document libraries and user accounts. Your results are not mixed with another organisation's documents. This applies on the pilot server too, which can run other organisations' pilots, each in its own separate workspace. Documents, chats and recordings are processed on the server that runs your workspace, by AI models that run on that server, apart from what the outside connections described below send out.
• Who can access it. Each person signs in with their own account, and your administrator decides who can use which assistants and features. In a workspace we host for you, and on the pilot server, OpenExpertsAI's administrators can access the server to maintain and support it. Hosted workspaces are reached over HTTPS through a tunnel service; when yours is reached through Cloudflare, Cloudflare's network decrypts and re-encrypts the traffic on its way to the server, possibly outside Vietnam. On a server in your own office, your organisation controls access.
• What can leave the server. Web search and outside AI are switches your administrator controls, and we agree with you during setup which ones are on. With web search on, search queries made from a user's question are sent to a web search service to fetch public pages. Documents are not sent, but a query can contain words from the question, such as a name. Outside AI is off unless your lab chooses it, and only the accounts your administrator allows can use it. When it is used, the conversation is sent to the outside provider, including any document passages in it. Before it goes out, names, ID numbers, phone numbers and other personal details are automatically replaced with placeholders, and the real values are put back into the answer on the server. This masking does not cover unpublished figures or results: if a message to outside AI contains them, they are sent as written. The masking catches common forms of personal details but is not perfect, so we agree rules with you on what staff may send.
• No training on your data. OpenExpertsAI does not use your data to train AI models. The models on the server do not learn from your documents or chats.

Read more about security and data handling

Where a person stays in charge

Answers can be wrong. That is why each answer points to its source: the researcher opens the passage or the data behind it and checks the figures before they are used.

A person reviews each masked figure and approves what is sent to a partner. Meeting notes stay drafts until someone edits and confirms them. The workspace helps with the work; it does not replace a researcher's judgement or sign-off.

Try it for a month

1. A short call about the work that takes your team the most time and the data that needs protecting.
2. We set up a private workspace for your lab, load your documents and add accounts for your team.
3. Your team uses it on real work for one month. During the pilot, the workspace runs on a server OpenExpertsAI operates.
4. You choose where it runs: a workspace we host for you, or a server in your own office.

If you stop after the pilot, we delete your workspace and data and confirm it in writing. See how a pilot works

See also

• For law firms
• For health clinics
• For research teams
• FAQ