Isomorphism 0.1.0 released

Isomorphism 0.1.0 is out. It is the first Python package in this series: isomorphism on PyPI, MIT-licensed, Python 3.12 or newer, no dependencies.

Why it exists

A language model does not need a client's AHV number or IBAN to summarise a letter. Isomorphism replaces such values with placeholders like [AHV_1] before the text leaves your system, and restore() puts the real values back into the answer. The same value always gets the same placeholder, so the model can still follow who is meant.

Everything runs locally: no network calls, no telemetry, no model download. The mapping with the real values is never logged or stored by the package.

What is in 0.1.0

• AHV numbers (prefix 756, EAN-13 check digit), IBANs of Switzerland and Liechtenstein (mod 97), UID numbers (mod 11), Swiss phone numbers in their usual spellings, and email addresses.
• Numbers with a wrong check digit are still found, with lower confidence. A typo in a document is still personal data.
• anonymize(text, lang=None, keep=()), restore(text, mapping) and detect(text).
• An audit record per hit with kind, position, confidence, recogniser and check result, never the value.
• keep=[…] for values that must stay, such as an authority's address.
• Hit rate and false alarms per kind, measured on invented Swiss texts in German, French, Italian and English, in the README.

Not in 0.1.0

Names, postal addresses, company names, dates of birth, account and card numbers, and restoring a streamed answer come in the next version. Until then names and addresses stay in the text. No tool finds all personal data, so treat Isomorphism as one protective layer.

Install

pip install isomorphism

from isomorphism import anonymize, restore

text = "AHV 756.1234.5678.97, IBAN CH93 0076 2011 6238 5295 7, Tel. 071 123 45 67"
safe, mapping = anonymize(text, lang="de")
# 'AHV [AHV_1], IBAN [IBAN_1], Tel. [PHONE_1]'

answer = call_your_model(safe)
print(restore(answer, mapping))

Try it in the live demo: the real package runs in your browser. The reference is at packages.sweber.dev/isomorphism/docs.