Add a language

By default pii-airlock looks for names in English and French. It checks every configured language on every text and combines the results, so mixed text is handled.

Add German as an example

1. Create .pii-airlock.yaml in your project:

   languages: [en, fr, de]
   models:
     de: de_core_news_lg

2. Download the models and check:

   pii-airlock download-models
   pii-airlock doctor

   download-models picks up the new language from your file. If it reports that your environment has no pip (common with pipx), it prints the exact pipx inject commands to run instead.

More models: spaCy models list.

Many languages with one model (GLiNER)

If you work in many languages, one multilingual model can replace one model per language.

pipx inject pii-airlock 'pii-airlock[gliner]'

ner_backend: gliner

pii-airlock download-models

	Standard spaCy models	GLiNER
Download size	small	about 1.1 GB, plus PyTorch
Speed per message	about 6 ms	about 40 ms
Names found correctly	good	better, with far fewer false alarms

Emails, phone numbers, IDs and secrets are found by patterns in both modes.