Your conversations and what is sent

The full and binding account is section 3.18 of the privacy policy. This page is the short version of it, in the same facts. Where the two seem to differ, the privacy policy is right and this page is a bug.

Your conversations belong to your account#

• They are stored in Nemi, on your own account: the messages, the actions taken and their results.
• Only you can read them. The other members of your workspace cannot, even the owner, and even for a conversation about work everybody shares.
• You can rename a conversation or delete it, from the list at the top of the panel.
• Deleting your account deletes them with it. See Closing your account.

What leaves our servers#

Nemi AI runs on a language model we do not host ourselves, so using it sends part of your content to a company outside the EU that runs it. That happens only when you open the panel and send a message. What is sent is:

• The message you typed, and the rest of that conversation.
• The results of the actions it took for you. Those results contain your data: file and folder names, the text of a document, the cells of a spreadsheet, the questions and answers of a form, the title and time of an event, the details of a photo.
• What you have open and what you have selected when you send the message: the title of the document, spreadsheet, canvas or form in front of you, the folder you are in, and the name, type and size of the files you have ticked or are previewing. Your browser sends only the identifiers, and each one is looked up against your own access before anything goes out. It is a record of what you were doing at that moment, so it is worth knowing that it travels with the message.
• Any image you attach or ask it to look at, made smaller and re-encoded first.
• Your display name and the name of the workspace you have open, so the answer makes sense.

A file's contents are sent only when you ask it to read that file. It does not read your storage on its own. Section 6.1 of the privacy policy names the two providers involved and Annex 3 of the Data Processing Agreement lists them as well.

What is never sent#

We do not train on any of it#

We do not use your content to create, train or improve AI or machine learning models, in raw or in derived form, and that includes what you say to the assistant. Every request we send on your behalf carries an instruction that it may not be used for training and may only be served by a host that does not retain it.

What we keep besides the conversation#

One record per request, holding which model answered, how much it used and what it cost. That is how the weekly allowance is counted and how we notice abuse. It holds no message text. See The weekly allowance.