What DataAttendant is, and why we built it
DataAttendant prepares CSV and Excel files for AI tools and other outside recipients. It replaces or removes confidential values in your browser. This page explains the problem, what the tool does and does not do, and the choices behind it.
Are you sharing sensitive information with an AI tool? Many people now work with an AI assistant next to their spreadsheets. An analyst wants a pivot summary explained. A finance team wants an odd pattern in last quarter's invoices found. A support lead wants to know what customers complain about. The data behind these questions is often data that should not leave the building: names, email addresses, account numbers, and amounts that identify a deal.
DataAttendant is a browser-based tool for the step in between. You open a CSV or Excel file. The tool shows you what looks sensitive. You decide what happens to each kind of value. Then you download a protected copy to share. The work happens in your browser, on your computer. It is also for private individuals: for example, someone who wants an AI assistant to look at a bank statement without sharing an account number or the names of the people they paid.
Key terms
- Finding
- a column, or a part of a text, where DataAttendant has found something that looks sensitive, such as an email address or a person's name. A finding is a proposal. Nothing changes until you accept it.
- Treatment
- what happens to the values of a column in the protected copy: removed, masked, replaced with a stand-in, encrypted, rounded, and so on.
- Stand-in
- a replacement value, such as
PERSON_001in place of a name. The same name always gets the same stand-in, so counts and groups still work. - Protection Plan
- your accepted decisions, saved so you can apply them to the next file of the same kind.
- Restore
- putting the original values back, on your computer, into text or a file that contains the stand-ins.
What is the problem?
There are three ways people deal with it today, and each has a cost.
They paste the data anyway. It is quick and it works. It also puts real customers' or colleagues' details into a service run by another company, under that company's terms. See what data should not go into an AI tool.
They clean the file by hand. Delete the name column, search and replace a few values, hope nothing is left. This is slow, and the misses are the dangerous part: names inside a notes column, a second sheet nobody looked at, a comment on a cell. See how to remove personal details from a spreadsheet and hidden data in Excel files.
They do not use the tool. The analysis the assistant could have done does not happen. Or it happens on a made-up sample that behaves differently from the real data.
Enterprise tools for this problem exist. They are built for data teams with databases, budgets, and implementation projects. A person with one spreadsheet and a deadline rarely has access to one.
What does DataAttendant do?
In order:
- You open a file. CSV and Excel files are read in your browser.
- DataAttendant looks for sensitive values and marks the columns where it finds them: names, email addresses, phone numbers, payment card and bank account numbers, dates of birth, identifiers, and parts of free text.
- You decide. You accept, reject, or change each finding, and choose a treatment for each column.
- You preview the result before and after, side by side if you prefer.
- It lists what remains exposed in the protected copy, and you acknowledge each warning.
- You download the protected copy and a record of what was done, which holds counts and fingerprints and not the values.
- You can save your decisions as a plan for the next file.
- You can restore original values in the assistant's answer, on your computer.
The page how DataAttendant works goes through each step in more detail.
Why does it work this way?
Why does the processing happen on your computer?
Because the aim of protecting a file is to avoid handing it to another party. A service that asks you to upload the original file first has moved the problem. DataAttendant reads and changes your file in your browser. The contents of your file, the values in it, and the names of your plans are not sent to us.
What is sent to our servers
You need a signed-in account to download a protected copy, and each export is checked with our server first. That check sends the account, a random job number, the number of records, and the time. It does not send the file's name, the column names, any value, or any part of your plans. Opening a file, reviewing it, previewing it, and restoring are not counted and need no account.
You can test this claim yourself. Your browser's network panel shows every request the page makes.
Why do you decide each treatment?
Because the right treatment depends on what you want to do with the file, and a tool cannot know that. A name column you want counted needs a stand-in. The same column in a file where names are irrelevant should be removed. An amount you want summed should not be masked. DataAttendant proposes, shows the effect of each treatment on an example, and leaves the choice with you. Nothing happens until you accept it.
Why stand-ins that can be put back?
Because a protected answer is only useful if you can read it in terms of real people. If the assistant tells you that PERSON_012 is behind half of your refunds, you need to know who that is. The mapping that makes this possible stays with you, protected by a passphrase that only you hold. See putting the original values back after an AI answers.
Why start with CSV and Excel?
Because that is where most day-to-day data work happens, and because a file is a natural unit to review: you can see what is in it and check what is left. Other sources may follow.
What does DataAttendant not do?
It also does not claim to be a complete data protection program, a certification, or a substitute for your data protection officer or your lawyer.
Is it right for your file?
Sources
- Regulation (EU) 2016/679 (GDPR), Article 4(1) (personal data) and Article 4(5) (pseudonymization).