Skip to main content
ZeroEval supports opt-in source-side PII redaction in the Python and TypeScript SDKs. When enabled, sensitive values are redacted before spans are buffered, logged, or sent by the SDK.
This is an SDK-side feature. ZeroEval does not rely on backend-side redaction for this behavior.

​
Enable redaction

import zeroeval as ze

ze.init(
    api_key="YOUR_API_KEY",
    redaction={"enabled": True},
)
You can also enable the default redaction toggle with: 
export ZEROEVAL_REDACT_PII=true
Python uses snake_case nested keys such as redact_inputs, redact_session_names, sensitive_keys, and custom_patterns. TypeScript uses the camelCase equivalents redactInputs, redactSessionNames, sensitiveKeys, and customPatterns.

​
What gets redacted

By default, the SDKs redact sensitive values found in:
  • inputs and outputs
  • attributes
  • error messages and stacks
  • session names
  • tag values
Built-in detectors cover:
  • email addresses
  • phone numbers
  • SSN-style identifiers
  • PAN / credit card numbers
  • bearer tokens, JWTs, and common API key formats
  • cookie and authorization header values
  • IP addresses
Key-based detection also force-redacts common sensitive fields such as email, phone, password, token, authorization, cookie, and api_key / apiKey.

​
What stays intact

Redaction is meant to preserve trace structure. The SDKs keep:
  • span names
  • trace IDs and span IDs
  • timing and status fields
  • token counts, model/provider metadata, and cost metadata unless the value itself is sensitive

​
Placeholder behavior

Sensitive values are replaced with stable placeholders inside a single trace: 
[email protected] -> [REDACTED_EMAIL_A]
[email protected]   -> [REDACTED_EMAIL_B]
[email protected] -> [REDACTED_EMAIL_A]
  • the same normalized sensitive value in one trace gets the same placeholder
  • different values in the same trace get different placeholders
  • placeholder assignment resets per trace
  • matching is exact after normalization only
  • this is not fuzzy identity resolution
That means repeated references across parent and child spans stay joinable without preserving the original value.

​
Normalization and limitations

TypeNormalization used for placeholder reuse
EmailTrim + lowercase
PhoneDigits only
SSN / PANDigits only
IPCanonical or lowercase string as implemented by each SDK
SecretsExact trimmed string
Important limitations:
  • there is no reversible backend token vault
  • there is no de-anonymization support
  • there is no fuzzy entity resolution across semantically related strings
  • bypassing SDK capture paths bypasses this protection

​
Examples

The SDK repositories include runnable examples that match the implemented behavior:
From zeroeval-sdk/examples/pii_redaction.py:
import zeroeval as ze

ze.init(
    api_key="sk_ze_demo_local",
    redaction={"enabled": True},
)

with ze.span(
    name="pii-redaction-demo",
    session={"id": "[email protected]", "name": "Alice [email protected]"},
    tags={"customer_email": "[email protected]"},
) as span:
    span.set_io(
        input_data={"email": "[email protected]", "phone": "+1 (415) 555-1234"},
        output_data={"result": "Reach [email protected]"},
    )
    span.set_error(
        code="ValueError",
        message="Failed for [email protected] with Bearer secret-demo-token",
    )
In both SDKs, repeated exact values such as the same email address will reuse the same placeholder within the trace.