Upload data

Your exports are read in this browser, on your own computer. Nothing is sent until you send it.

Select
Filter
Upload

Drag your data here

Or click here to select files, or

Overview

Data included	Included	Excluded

Read what is in the file

Metadata in the file

What happens to your data

Where it is read

Your export files are read in this browser tab, on your own computer, and nothing is uploaded while you read. Closing the tab destroys everything the tool has read.

What is replaced

Everything you wrote is labelled SUBJECT and everything the other person wrote is labelled OTHER. Only the sender label is replaced. If a message says a name out loud, that text is kept exactly as written, because it is part of what was said.

What is sent

Conversations, their text, and voice notes. Photos and video are counted and left on your computer: they are not part of what the study reads. You see the counts before you send anything.

Volume, in the table before you send, counts threads, messages, and message text — characters, and tokens estimated from that text (within about ten per cent of a real tokenizer). Message text is the only free text in the file; the layout around it weighs about six times as much, and voice notes travel as files.

The file also holds a log of how it was made: counts, the layout of each export format, and any error, with no names, no message text and no file names. It lets us fix a format that fails. The Debug log button shows you the same log.

How exports arrive

Facebook and Instagram write a large export as several zip files, most of which hold only media; WhatsApp writes one chat per file. Choose all of them, or the folder they are in — they are read together as one donation, and the parts without conversations cost nothing.

What leaves your computer

Only the file you build. It is streamed from this tab straight to the university’s storage, and the tool always shows you what is in it first.

Keeping your data private

We are two master students. Our research project asks whether a language model, given enough of someone’s chats with other people, can answer survey questions the way that person would.

Answering that takes real conversations, which is a lot to ask of anyone. So we made this interface for full transparency. Instead of asking you to hand over your export, this runs on your own computer and we never receive the archive. You can read everything in a file before sending it to us.