Markdown cleaner

What does this tool do?

You say what should go, and the tool takes it out of your Markdown files. In practice that is usually something nobody outside the team should see: a real name in a meeting note, a customer number in a technical doc, an e-mail address in a changelog. You enter the terms, the tool finds them in every uploaded file, and it replaces, redacts or deletes them.

You decide three things:

1. What gets searched. A list of terms, one or more pattern templates (e-mail, phone, IBAN, all numbers, URLs), and optionally your own search pattern.
2. What happens to matches. Remove deletes them and closes the gap. Redact replaces them with a placeholder such as ████. Replace swaps them for text you provide, like [Customer].
3. How precisely it searches. Whole word, exact with case, or partial inside the word.

Then you see a before-and-after preview of the first file and a summary of how many matches were removed across how many files. You download the cleaned files one by one or all together as a ZIP.

Why local instead of a server?

Markdown files often carry exactly the content you do not want to hand over: notes with real names, jottings with credentials, exports from a knowledge tool full of internal IDs. Many online tools for “replace text” or “redact data” upload the file to a server to process it. For sensitive content that is the very problem you were trying to solve.

This tool has no upload. The page is served over static hosting, all of the code runs in your browser tab, and there is no endpoint that could accept file contents. The term list stays local too. It can hold a real person’s name, and that has no business on someone else’s server, so it is never written to browser storage and disappears as soon as you close the tab.

Why clean Markdown before an AI chat?

More and more people paste notes, code snippets or whole documents into an AI assistant to summarise or rework them. That is handy, but everything you paste travels to an outside service. If there is a customer name, a contract number or a private address in it, you hand that over too.

This is where the tool fits in. You clean the Markdown file once, locally, swap the sensitive parts for neutral placeholders, and only then paste the cleaned version into the chat. The assistant sees [Person 1] instead of the real name and can still summarise, rewrite or answer questions. Replace mode is built for exactly this: consistent placeholders instead of black bars.

How do templates handle recurring patterns?

Some things you do not type one by one, because they show up in many shapes. An e-mail address might read anna.miller@company.com or a.m+invoice@sub.company.co.uk. That is what the templates are for. One click on the chip, and the tool finds every e-mail address, phone number, IBAN, number or URL in the text. The patterns are vetted and built to catch the common formats without grabbing random fragments of words.

If you need more control, open the field for your own pattern and enter a regular expression, for example ID-\d{4} for internal record numbers. If the pattern is invalid, you get a notice instead of a crash, and the rest of the cleaning runs as normal.

How does the word-boundary detection work?

In Whole word mode, Berlin should match only the word Berlin, not Berliner. Behind the scenes the tool bounds each match with real Unicode word boundaries. That sounds technical, but it has a very practical effect for non-English text: accented letters and umlauts like ä, ö, ü and ß count as ordinary letters. Müller is recognised as one word, and Grüße is not cut after the ß by accident. Classic find-and-replace functions often only know the English alphabet and make exactly this mistake.

There are three precision levels. Whole word is the safe default and ignores case. Exact matches only the spelling you typed, including its case. Partial searches everywhere, even in the middle of longer words, and is useful when a term shows up in many inflected forms.

Which files are supported?

Accepted inputs:

• Markdown files with the .md or .markdown extension
• 1 to 100 files at a time
• Up to 20 MB per file

Output:

• A cleaned .md version of each file with a -cleaned suffix
• Single download or everything together as a ZIP archive
• A before-and-after preview and a match summary per run

Deliberately out of scope:

• PDF or Word redaction — there are dedicated tools for that, Markdown stays the focus
• Automatic name detection without your input — you decide what goes, which keeps the result predictable

Frequently Asked Questions

How do I remove confidential data from a file before giving it to an AI?

Upload the Markdown file here, type the sensitive terms or switch on a template, and click Clean. The tool hands you back a clean file that you can then safely paste into an AI chat. Nothing is uploaded in the process.

Are my files uploaded when I clean them?

No. Reading, searching and cleaning all happen in your browser tab. There is no server that could accept the contents, and the search terms are not stored either.

Can the removed text be recovered later?

No. Markdown is plain text, so removing or redacting deletes the character itself rather than covering it. Nothing hidden is left in the downloaded file.

Does it handle accented characters and German umlauts?

Yes. Word boundaries use the full Unicode alphabet, so ä, ö, ü and ß work correctly. Müller is recognised cleanly as a whole word and Grüße is not split.

Which Tools Are Related?

• CSV cleaner — fix encoding, delimiter and German number format in CSV exports automatically.
• Remove metadata — delete hidden file metadata from images and PDFs locally.
• Markdown to PDF — turn the cleaned Markdown file into a finished PDF.