Platform / PII Redaction

Entity-level PII redaction for LLM calls

Detect and replace sensitive identifiers before they reach the upstream model. Replaceable mask tokens preserve reasoning quality without exposing raw personal data.

Request Access View docs

How redaction works

Deterministic redaction — same entity, same mask token, every time — runs on every prompt before it exits your application's network boundary.

Before and after diagram showing PII entity detection and redaction masking in a prompt string

Supported entity types

30+ built-in entity types. Add custom patterns via YAML config.

PII_NAME

PII_EMAIL

PII_PHONE

PII_SSN

PII_DOB

PII_ADDRESS

PII_ACCT_NUM

PII_CREDIT_CARD

PII_IP_ADDR

PII_NPI

PII_MRN

PII_DIAGNOSIS

FIN_CUSIP

FIN_ISIN

FIN_AUM_VALUE

CUSTOM_PATTERN

Policy configuration

Declare redaction rules in YAML. Version-control your policy alongside your application code.

Per-project PII/PHI/PCI entity lists with vertical-specific defaults
Custom regex patterns for proprietary identifier formats (internal client IDs, deal reference numbers, case numbers)
Replacement modes: mask token (default) / hash / suppress
Context-aware detection mode: healthcare (HIPAA Safe Harbor), financial (FIN_ACCT_NUM + CUSIP + ISIN), legal (CLIENT_NAME + DEAL_TERMS)
Non-determinism risk controls: deterministic mask assignment prevents re-identification across correlated calls

policy.yaml

redaction:
  enabled: true
  entities:
    - PII_NAME
    - PII_EMAIL
    - PII_SSN
    - PII_ACCT_NUM
    - PII_MRN
  mode: mask_token
  custom_patterns:
    - name: INTERNAL_CLIENT_ID
      regex: "CLT-[0-9]{6}"

Start redacting in under an hour.

Private beta — connect to your LLM stack and configure your first policy in one session with Kevin.

Request Access Read quickstart