Skip to content
Nappai Documentation

Data Anonymizer

Data Anonymizer finds personal data inside a text and replaces it with dummy values. For example, “Laura Gómez” becomes another made-up name and her phone number becomes a different number. It also keeps the mapping between each real value and its substitute, so you can later restore the original text with Data Deanonymizer.

This lets you send information to an AI model or an external service without exposing your customers’ data. The language of the text is detected automatically.

When to use it

  • Analyse customer emails or tickets with AI without sending their names, emails or phone numbers.
  • Share real conversation examples with an external provider without personal data.
  • Prepare internal documents (contracts, case files) before an agent processes them.
  • Follow your data protection policies in flows that handle sensitive information.

Configuration

FieldWhat it does
DataThe records containing the text you want to anonymise.
Source data input keyName of the field in each record that holds the text. Default: text.
Analyze FieldsTypes of data to find and replace: PERSON (people), ORGANIZATION (companies and organisations), LOCATION (places), DATE_TIME (dates and times), EMAIL_ADDRESS (emails), PHONE_NUMBER (phone numbers), IBAN (bank accounts), DNI (Spanish ID numbers) and CREDIT_CARD (cards). Default: people, organisations, emails and phone numbers.
NER Model NameHow the data is detected. spacy (default) uses the platform’s built-in recognition. gemma3:12b uses an AI model to locate it.
Remark AnonymizationWith the gemma3:12b option, wraps each replaced value in asterisks (***value***) so you can see what was changed. Off by default.

Advanced options

FieldWhat it does
Model SizeSize of the built-in recognition (small, medium or large). A larger one usually detects better but takes longer. Default: medium. Only applies with spacy.
Spacy Ignore EntitiesTypes of elements the built-in recognition should ignore, to avoid unnecessary replacements. Default: MISC, WORK_OF_ART and PRODUCT.

Output

The Data output returns the same records with two new fields:

  • text_anonymized: the text with personal data replaced.
  • anonymizer_mapping: the mapping between each real value and its dummy value. Keep it if you will need to restore the original text.

Example

You want an agent to classify customer complaints without the model seeing their data.

  1. Connect the complaints to Data.
  2. Keep people, emails and phone numbers in Analyze Fields, and add IBAN if customers often include their account number.
  3. Send the text_anonymized field to the agent.
  4. If the agent drafts a reply, use Data Deanonymizer with the anonymizer_mapping to put the real data back before sending it.

Good practices

  • Check a few results before using the flow with real data: no detection system is perfect.
  • Enable only the data types you need. Enabling LOCATION or DATE_TIME may replace information that is useful for the analysis.