Pii Pattern Redactor
Pricing
$0.01 / actor start
Pii Pattern Redactor
Pricing
$0.01 / actor start
Rating
0.0
(0)
Developer
Eichan Jolly
Maintained by CommunityActor stats
0
Bookmarked
2
Total users
1
Monthly active users
3 days ago
Last modified
Categories
Features
Share
Actor slug: pii-pattern-redactor
Store category: AI
Runtime: Python 3.12 standard library only; no Apify SDK, login, external LLM, or third-party AI service.
What it does
Take one user-supplied text string and replace detected patterns with [EMAIL], [PHONE], or [CARD]. Emit one Dataset item with redacted text and counts only; the findings array never contains matched values.
Who it is for
Developers who want a quick local, deterministic first-pass filter for logs, support content, prompts, or sample data before further handling. Do not use this as a standalone privacy/security control.
Method
- EMAIL: a conservative, ASCII-oriented email-address regular expression.
- CARD: 13–19 ASCII digits, either contiguous or separated by single spaces or hyphens. Only candidates that pass the Luhn checksum are replaced. Unsuccessful card candidates are excluded from partial phone matches.
- PHONE: plausible phone-like groupings (7–15 digits including optional country code, punctuation and spaces), plus unformatted 10–12-digit sequences. Fixed-group formatted phone boundaries are checked first so two phone numbers separated by a space are not accidentally treated as one number or a rejected card candidate. Pattern shape and digit length are checked; no country-specific telephone validation occurs.
- Match priority is EMAIL → CARD → PHONE, then matching spans are replaced left-to-right without overlaps. Every run is deterministic.
findingscounts are orderedEMAIL,PHONE,CARD, and only nonzero categories are present. No raw values, offsets, or snippets are placed in findings or terminal logs.
Input
Required text (string): 1–100,000 Unicode code points; whitespace-only strings are rejected. The prefilled Actor input schema contains the working sample below.
Output fields
redactedText (string or null on error), findings (list of { "type": "EMAIL|PHONE|CARD", "count": integer } objects, never raw values), totalRedactions (integer), status (success or error), and error (null or readable message). No matches is a successful item with the original text, findings: [], and totalRedactions: 0.
Limits, false positives and safety warning
This is pattern matching. It can miss real personal data and can flag look-alikes. It is NOT a compliance, privacy, or safety guarantee. Do not assume the result is anonymized, de-identified, GDPR/CCPA/HIPAA-compliant, PCI-safe, or safe to publish.
- Does not recognize names, addresses, IDs, free-form disclosures, many international/local phone formats, non-ASCII emails, or obfuscated values.
- Luhn is a checksum, not proof that a number is an issued card. Conversely, an incorrectly entered card number fails Luhn and is not labeled as CARD.
- Phone-like sequences can be mistaken for order numbers. Unformatted telephone patterns have conservative size limits; numbers outside them may remain.
- If no pattern matches, input text is returned unchanged, even if it still contains sensitive information. Manually review outputs before sharing.
- There is no content fetching or external AI API. On Apify, input/output do traverse Apify platform storage via its native REST API; this Actor does not send content to a separate service.
- Invalid input yields one
status: "error"Dataset item and a nonzero exit after saving it.
Working input example (also the exact schema prefill)
{"text": "Email me at sam@example.org or call +1 415-555-2671. Test card: 4111 1111 1111 1111."}
Actual output example (from script-mode smoke run)
{"redactedText": "Email me at [EMAIL] or call [PHONE]. Test card: [CARD].","findings": [{"type": "EMAIL","count": 1},{"type": "PHONE","count": 1},{"type": "CARD","count": 1}],"totalRedactions": 3,"status": "success","error": null}
How to run locally
From this Actor folder with Python 3.12+:
# Bash/macOS/Linux: reuse the proven sample input.ACTOR_INPUT_JSON="$(cat tests/sample_input.json)" python src/main.py
Or create storage/key_value_stores/default/INPUT.json containing the input JSON, then run python src/main.py. Local Dataset items appear in storage/datasets/default/000000001.json (with increasing sequence numbers). You can override the storage root with APIFY_LOCAL_STORAGE_DIR.
Apify platform behavior (no SDK)
The Actor selects input in this order:
ACTOR_INPUT_JSON(when set, even if it contains invalid JSON; invalid input is reported instead of silently falling through).- When
APIFY_DEFAULT_KEY_VALUE_STORE_IDexists:GET https://api.apify.com/v2/key-value-stores/{id}/records/INPUTwithAuthorization: Bearer {APIFY_TOKEN}. - Otherwise: local
./storage/key_value_stores/default/INPUT.json(or theAPIFY_LOCAL_STORAGE_DIRequivalent).
On the platform, one result is appended to POST https://api.apify.com/v2/datasets/{APIFY_DEFAULT_DATASET_ID}/items using the runtime token. The Actor does not require customers to provide an API key or log in to a third-party service. Both storage requests have a 20-second timeout; network, HTTP, and TLS certificate problems return readable errors without tracebacks. No custom API base URL validation is imposed.
Reproducible verification
python -m unittest discover -s tests -vruns included automated tests, including direct subprocess script mode and local REST mocks.python tests/run_smoke.pyre-runs the actualpython src/main.pywith the schema prefill, saves a local Dataset record, and regeneratestests/smoke_evidence.md,tests/smoke_stdout.txt, andtests/smoke_dataset_item.json.python tests/run_regression_smoke.pyruns the corrected two-phone case and the mixed-text case via real subprocess execution and saves their Dataset records and terminal outputs undertests/regression_smoke/.tests/unittest_terminal.txt,tests/smoke_evidence.md, andtests/regression_smoke/evidence.mdcontain captured outputs from the delivery verification. Live Docker builds and Apify-platform deployment were not executed.
Store category and suggested pricing
Category: AI (data preparation utility). Suggested pricing: USD $0.005 per completed run; suggestion only. It is recorded in store_metadata.json. No publishing or billing setup was performed.