Clean a file or some text

Use this when you want to hide personal data by hand, for example before pasting into a website chat that pii-airlock cannot sit in front of.

The three-step routine

1. Hide

pii-airlock scrub notes.txt --map notes.pii-map.json -o notes.clean.txt

• notes.clean.txt is safe to share. It contains placeholders.
• notes.pii-map.json is the key to undo it. Keep it private. See the warning below.

You can also pipe text in:

echo "Contact Jean Dupont at jean@acme.fr" | pii-airlock scrub --map m.pii-map.json

Contact <PERSON_558644> at <EMAIL_ADDRESS_094087>

2. Use the clean text with your AI

Paste notes.clean.txt into the AI. It answers using the placeholders.

3. Restore

Save the answer to a file, say answer.txt, then:

pii-airlock restore answer.txt --map notes.pii-map.json -o answer.final.txt

The placeholders are replaced by the real names and emails.

The map file contains the real data

Never email it, upload it or commit it to Git. Delete it when you have finished restoring. If you do not need to restore, use --no-map, which makes the cleanup one-way.

Which files work?

Type	Works	Notes
Plain text, CSV, JSON	Always
Word .docx	Needs the docx extra	Changed paragraphs lose inline formatting such as a bold word in the middle.
PDF	Needs the pdf extra	Text only. You get cleaned text, not a new PDF.

pii-airlock scrub report.csv -o report.clean.csv
pii-airlock scrub contract.docx -o contract.clean.docx
pii-airlock scrub scan.pdf > scan.clean.txt

Install the extras from Install, all options.

See what it would hide first

pii-airlock detect notes.txt

PERSON           0.85  [9:21]    'Jean Dupont'
EMAIL_ADDRESS    0.99  [24:38]   'jean@acme.fr'

Nothing is changed. The number is the confidence from 0 to 1.

Useful options

Option	What it does
--no-map	One-way cleanup. You cannot restore afterwards.
--encrypt-map	Encrypts the map so it only opens on this computer.
--lang fr,en	Choose the languages to look for.
--threshold 0.7	Be pickier. Fewer false alarms, more risk of a miss.