Supported Languages

Agent Mask uses language-specific detection models for supported languages. Pass a two-letter ISO code via the language parameter to select the language model for a mask() call.

Supported codes

Code	Language
en	English (default)
es	Spanish
pt	Portuguese
de	German
fr	French
nl	Dutch
it	Italian
no	Norwegian
pl	Polish
ro	Romanian
hr	Croatian
da	Danish
fi	Finnish
el	Greek
uk	Ukrainian
cs	Czech
ca	Catalan
sv	Swedish

Deployment availability

The language parameter selects among the models included in your Agent Mask deployment. It does not download or add models at runtime.

For self-hosted Docker deployments, language availability is selected by image tag:

Tag	Language coverage
latest, en-latest, X.Y.Z-en	English
<lang>-latest, X.Y.Z-<lang>	One supported language, such as fr-latest
<lang>-<lang>-latest, X.Y.Z-<lang>-<lang>	A selected language subset, such as en-fr-latest
multi-latest, X.Y.Z-multi	All supported languages

For published Docker images, language availability is baked into the selected image tag. Set STARTUP_LANGUAGES to preload languages already included in the image at container startup; it does not add model files or enable languages absent from that image.

Usage

SELECT app_public.mask(
  'Maria Garcia-Lopez, 45 años, nacida el 06/30/1979',
  OBJECT_CONSTRUCT('language', 'es')
);

Pattern-based entities and language

Most pattern-based types (EMAIL_ADDRESS, PHONE_NUMBER, CREDIT_CARD, US_SSN, MEDICARE_ID, NPI, IBAN_CODE) detect correctly regardless of the language setting.

Some label-aware pattern recognizers use the selected language for context words around the identifier. These include ACCOUNT_NUMBER, DEVICE_IDENTIFIER, LICENSE_PLATE, MEDICAL_LICENSE, MEDICAL_RECORD_NUMBER, VEHICLE_VIN, AGE_OVER_89, and ZIP_CODE.

The language parameter mainly affects AI-detected types (PERSON, LOCATION, ORGANIZATION, DATE_TIME, NRP) and label-aware pattern context.

Documents and OCR

Document processing uses the same language setting as text masking. For scanned PDFs and image files, Agent Mask also passes the selected language to OCR when that OCR language is supported. The setting selects detection and OCR behavior; it does not translate the source text.

Mixing languages in a batch

Run separate mask() calls per language. A single call processes all input with one language model.

Detection quality can vary by language, document quality, OCR quality, entity type, and input format. Review outputs before using them for external sharing, regulated disclosure, or public release.