Krakt Blackt: PDF redaction that actually deletes the text

A black box drawn over a PDF hides nothing: select, copy, paste, and the "redacted" SSN is right there. Blackt deletes the text underneath, then re-opens the result and checks. About $0.01 per file.

• Real removal: PyMuPDF redaction annotations + apply_redactions. Glyphs are deleted from the content stream, the file is fully rewritten (no incremental-save leftovers).
• Batch it: 1 PDF or 1,000. URLs, an upload, or key-value store records.
• Built-in PII presets: emails, phone numbers, US SSNs, credit cards (Luhn-checked), IBANs (mod-97 checked).
• Your own rules: exact terms/phrases and Python regexes. Optional page ranges like 1-3,5,10-.
• Proof, not promises: every file gets a removal report: counts by type, pages hit, and a verification pass that re-extracts the text and re-runs every search. verified: true means zero hits.
• Metadata scrubbed by default (title, author, keywords, XMP).
• Pay for results: failed downloads and broken PDFs are free.

What it won't do (read this)

• No OCR. Scanned or image-only pages have no text to find, so nothing is redacted there. We flag them in imageOnlyPages so you know to check by hand.
• Detection is pattern-based. Presets catch common formats, not every format on earth. Names, addresses and free text need search terms or a regex.
• Text must be extractable. Weird font encodings or text drawn as vector paths can hide matches. Those show up as misses, not as errors, so spot-check sensitive files.
• Phrases wrapped across lines are found, but each word in the phrase is then removed wherever it appears on that page (we over-redact rather than leak).
• Password-protected PDFs are skipped. Annotations, form field values and attachments outside the matched areas are not touched.
• Verification proves the matched text no longer extracts. It is not a legal certification. Have a human review anything that goes to court.

Quick start

{
  "pdfUrls": ["https://example.com/contracts/acme.pdf"],
  "presets": ["emails", "phones", "ssns", "creditCards", "ibans"],
  "searchTerms": ["Project Nightingale", "Jane Doe"],
  "customRegexes": ["\\bEMP-\\d{6}\\b"],
  "pageRanges": "1-5"
}

Field	Default	What it does
pdfUrls	–	Direct links to PDFs
pdfUpload	–	Upload one PDF from the Console
kvStoreRecords	–	storeId/key (or key for this run's store)
presets	all five	emails, phones, ssns, creditCards, ibans
searchTerms	–	Exact words or phrases
customRegexes	–	Python regexes
caseSensitive	false	Case-sensitive term matching
pageRanges	all	e.g. 1-3,5,10-
boxColor	black	black, white, navy
replacementLabel	–	Text inside each box, e.g. [REDACTED]
redactImageAreas	true	Blank image pixels under boxes
scrubMetadata	true	Wipe document metadata
maxFiles / maxFileSizeMb	100 / 50	Safety caps

Output

Key-value store: 001-name-redacted.pdf and 001-name-report.json per file. Dataset: one row per file:

{
  "fileName": "hr-record.pdf", "status": "redacted", "verified": true,
  "totalRedactions": 36,
  "redactionsByType": { "emails": 6, "phones": 6, "ssns": 3, "creditCards": 6, "ibans": 6, "terms": 9 },
  "pagesRedacted": [{ "page": 1, "redactions": 12 }],
  "verification": { "passed": true, "removedStringsStillExtractable": 0, "residualMatchesByType": {} },
  "imageOnlyPages": [], "redactedPdfUrl": "https://api.apify.com/v2/key-value-stores/.../records/001-hr-record-redacted.pdf"
}

status is redacted, no-matches (still a clean, metadata-scrubbed copy) or failed.

Pricing

$0.01 per processed file (pay per event). Failed files cost nothing. A run with no PDFs given redacts one public sample form (IRS W-9) for free (demo: true) so you can see the output. The report itself never contains the removed values.

Built by Krakt.