--detect-pii

Scan a file for columns that look like personal data and print the findings to stdout. Read-only.

octa --detect-pii FILE [--pii-sample N] [-f tsv|json|csv]

--pii-sample sets how many rows are sampled per column for pattern matching (default 500).

How it works

Each column is scored on two clues: its header (email, name, gender, country, birthdate, ip, ...) and its values (email, phone, IP, credit card, IBAN, SSN, date, postal-code shapes). A strong value pattern alone, or a matching header alone, reaches 0.5; together they score highest. So name, gender and country columns are caught from the header, while a plain number column like salary is left alone.

Output

One row per finding with these columns:

Column	Meaning
column	Column name
kind	email, phone, ip_address, credit_card, iban, ssn, name, gender, country, birth_date, postal_code, address
confidence	0..1 score (reported when >= 0.5)
by_name	Whether the header matched
value_match	Fraction of sampled values matching the kind's pattern

by_name and value_match show how the confidence was reached. Use it to decide which columns to mask with --anonymize.

Examples

octa --detect-pii customers.csv
octa --detect-pii customers.parquet --pii-sample 2000 -f json

See also

• Detect PII (GUI) and the detect_pii MCP tool.
• --anonymize: mask the columns you find.