Technology
A small window in the workspace shows in which project your assistant searched and which documents it found. The assistant itself does not get the names.
PrivaSearch has an activity window in the workspace. When you work with Claude, ChatGPT Business or Mistral Vibe and your workspace is open in the browser, you see at the bottom right in which project your assistant searched and which documents it found, with the page. One click on a name opens the project or document.
The page about the activity window has a one-minute video that shows it: the assistant's chat window on the left, the workspace on the right, and what happens in between when the assistant searches.
A connected assistant talks to PrivaSearch through MCP, the protocol that lets an assistant use tools from another service. What that is and how it works is explained in What is an MCP server?. The assistant asks, for example, which projects there are, or searches a project for a passage.
PrivaSearch deliberately returns no names. In practice a project name or a file name is often the name of a client, a patient or an employee: "File De Vries" or "Dismissal letter Jansen.pdf". That name is personal data, and the assistant does not need it to search well. The assistant therefore gets short references, such as "Project VTC7AQ" or "Document 9KD2LM", and the passages it reads are already pseudonymised. The difference between pseudonymising and anonymising is explained in Pseudonymising or anonymising.
That fits a basic rule of the GDPR: you process no more personal data than the purpose needs. The rule is called data minimisation and is in Article 5(1)(c) of the GDPR. How PrivaSearch works as a whole, from upload to answer, is on How PrivaSearch works.
References protect the names, but they are hard to read. Ask the assistant about "the tenancy agreement of that client in Utrecht" and it answers with a passage from "Document 9KD2LM" in "Project VTC7AQ". In the chat window you don't immediately know whether it had the right file.
Until now you had to work that out yourself: copy the reference, find the project in the workspace and open the document. The activity window does it for you. While the assistant searches, you see in your workspace, by name, which project and which documents it used. You check at a glance whether it was in the right file, and with one click you have the document open to check the answer.
The names do not go to the assistant. They only appear in your browser, in your own workspace, and only as far as you may see them now.
An example with fictitious data. Sanne works at an accounting firm and has connected Claude to her firm's workspace. She has two windows open: Claude on the left, the PrivaSearch workspace on the right.
Claude did not see the name of the client or of the file in that whole exchange. Sanne did, because she is a member of that project.
Each line in the window is one action of your assistant. At the top it says which assistant it was and at what time. Below that it says what it did:
| What your assistant did | What the window shows |
|---|---|
| It asked which projects there are | Looked at your projects |
| It looked at the documents of a project | Looked at the documents of, with the project as a link |
| It searched in a project | Searched in, with the project and the documents found with their page |
| It read a source | Read a source from, with the project and the document with its page |
For a search, a line holds at most five documents, each once, with the page where the passage was. A click on a project or document opens it in the workspace. The window itself never opens anything: only a click from you takes you somewhere.
The window shows the last five actions. A line disappears by itself after five minutes. The cross closes the window; it comes back as soon as your assistant does something again.
The window is meant to show where your assistant searched, not to read along. So it never contains:
For the window PrivaSearch also keeps only what the window shows: the kind of action, the project, the documents with their page and which assistant it was. After ten minutes that is gone again.
The names are also only looked up when the window shows them, with your rights at that moment. If you have just been removed from a project, or a project's retention period has passed, the project no longer appears in the window. If a document has been deleted or you may no longer see it, it drops out of the line.
The window only checks for something new while the workspace tab is visible. If the tab is in the background, it asks nothing. If nothing happens for three minutes, it stops checking. As soon as you return to the tab, it checks again straight away.
That is exactly the moment it matters: you ask your question in your assistant's chat window, then look in the workspace to see where it searched. On My profile, the AI training and the acceptance of the terms the window does not check.
The window is on by default for everyone. If you don't want it, open My profile and, under "What your assistant does", switch off "Show what my assistant does in the workspace".
Off means more than the window disappearing. PrivaSearch then also stops keeping track of what your assistant does, and anything still shown is gone at once. Your assistant keeps working, with the same references as always. If you switch it on again later, the window appears at your assistant's next action.
PrivaSearch already had an activity log. Among other things, it records when an assistant called PrivaSearch and which source passage it read. The activity window does not replace that log: the two have a different purpose and a different reader.
The activity log is the lasting record of what happened in the workspace, and you can search it later. An administrator of the organisation uses it to review usage, for example by filtering on a user or on calls by an assistant. A line in the log stays and is never changed afterwards.
The activity window is for yourself, and for now. It shows in plain language and by name what your own assistant did in the last few minutes, so you can place the answer in the chat window straight away. After five minutes a line is out of sight, and after ten minutes it is gone. If you switch the window off, nothing changes in the activity log: PrivaSearch simply keeps it.
In short: the log is the record afterwards, the window is a help while you work.
Everyone who works with a connected assistant gets something out of it. A few examples:
Which assistants you can connect and how is on AI chatbots and the GDPR. What PrivaSearch does for security is on Security.
No. The assistant gets the same short references as before. The names only appear in the window in your own workspace, and only you see them.
No. The window only shows what the assistant of your own account did. A colleague only sees their own assistant in their workspace.
No. For the window PrivaSearch keeps only the kind of action, the project, the documents with their page and which assistant it was. Your question and the text the assistant read are not in it.
A line disappears from the window by itself after five minutes, and after ten minutes it is also gone from PrivaSearch. The cross closes the window; it comes back as soon as your assistant does something again.
No. The window opens nothing by itself. Only a click from you on a project or document takes you there.
Yes. Switch off "Show what my assistant does in the workspace" on My profile. PrivaSearch then also stops keeping track of what your assistant does.
GreenPT is working on MCP with sign-in. What a connection with PrivaSearch would then offer: searching your client files with pseudonymised passages.
How retrieval-augmented generation works, how it differs from uploading and fine-tuning, why it reduces errors and which privacy questions to ask.
What the Model Context Protocol is, how a connection differs from uploading a file, what an assistant receives and which risks you should check.
We show it with fictitious client files from your sector: upload, detection, review, search and the connection with your AI assistant. Afterwards you'll know exactly what your team can use safely.
Book a demo →