Skip to main content

​
Goja vs. JMESPath for PII Redaction

When securing sensitive Personally Identifiable Information (PII) before allowing it to reach an LLM, the choice between JMESPath and Goja exclusively depends on whether you want to completely Drop the field or Mask the field.

​
When JMESPath is Better (Complete Drop)

If you simply want to ensure a credit_card_number or ssn node is unconditionally excluded from the final LLM Context, JMESPath is infinitely faster and safer. By writing a strict exclusive Allowlist (users[*].{name: name, id: id}), you guarantee that the PII fields are never dynamically forwarded securely.

​
When Goja JS is Better (Targeted Masking)

However, sometimes an LLM inherently needs to know partial identity parameters to make decisions, such as verifying the last 4 digits of an account number. JMESPath cannot split a string intelligently. Goja JS is required when you need to run Regular Expressions (Regex) across strings to replace characters deterministically. 
input.forEach(function(u) {
  if (u.account_number) {
    // Assuming format XXXX-XXXX-XXXX-1234
    u.account_number = u.account_number.replace(/\d{4}-\d{4}-\d{4}/g, "****-****-****");
  }
});

return input;
This transforms the payload natively and seamlessly. This ensures the LLM receives ****-****-****-1234.