Audit Mode

Audit mode lets you see what PII Rdakt AI would detect and anonymize, without actually modifying any requests. Use it to assess your application's PII exposure before enabling anonymization.

Enable Audit Mode

Via YAML

mode: audit

Via Code

from rdakt_ai import RdaktMiddleware

middleware = RdaktMiddleware(mode="audit")

Override at Middleware Level

from rdakt_ai import RdaktMiddleware, load_config

config = load_config("rdakt.yaml")  # mode: active
middleware = RdaktMiddleware(config=config, mode="audit")  # Override to audit

What Happens in Audit Mode

Step	Active Mode	Audit Mode
Detection	Runs	Runs
`on_entities_detected` callback	Fires	Fires
Anonymization	Replaces PII	Skipped
`on_anonymized` callback	Fires	Does not fire
Request to LLM	Anonymized text	Original text
Response deanonymization	Runs	Skipped

Logging Detected Entities

Combine audit mode with callbacks to build a PII inventory:

import json
from rdakt_ai import RdaktMiddleware, Entity

pii_log: list[dict] = []

def log_pii(entities: list[Entity], text: str) -> None:
    for entity in entities:
        pii_log.append({
            "type": entity.type,
            "value": entity.value,
            "position": f"{entity.start}-{entity.end}",
        })

middleware = RdaktMiddleware(
    mode="audit",
    on_entities_detected=log_pii,
)

# After running your test suite through the middleware:
print(json.dumps(pii_log, indent=2))

Workflow

Deploy with mode: audit — observe what PII is flowing to your LLM provider
Review the audit log — check for false positives, missing patterns
Add custom patterns if needed for domain-specific PII
Switch to mode: active — enable anonymization