piilite 0.3.0

• Sessions for RAG: shield.session() shares one set of placeholders across the question, passages, history and the restored answer. detectAll() / redactAll() scan several texts first.
• Known values are masked everywhere: once a value is found, later occurrences are masked even with no clue around them, including parts of names ("rahul" after "Rahul Sharma"), mapped to the same placeholder. Turn off with propagate: false.
• Placeholders ignore case and spacing again ("Rahul Sharma" = "rahul sharma", 9123456780 = +91 91234 56780); restore uses the best-capitalised spelling. matchCase: true keeps the 0.2.1 behaviour.
• Streaming restore: createRestoreStream() / restoreStream() hold back a placeholder split across chunks, in every form piilite restores.
• restore() understands spaces inside type names ("[PHONE IN 1]") and has fuzzy: "brackets" to leave bare words like EMAIL_1 alone.
• Company addresses are masked again by default (a home address can follow an employer's name); keepCompanyAddresses: true leaves them.
• Masking a question with 8 passages and 2 KB of history (about 9 KB) takes about 8 ms (p95 13 ms) on a 4-core server.

npm install piilite@0.3.0

piilite 0.2.1

• A company's own address ("Example Technologies Pvt Ltd, Sector 62, Noida") is no longer masked by default; the strict preset still masks it.
• Placeholders are case-exact: "RAHUL" and "Rahul" get their own placeholders, so restore() brings each back as written.
• New skipFakeNumbers option (default true). Set it to false, or use the strict preset, to also mask test-looking numbers such as 9876543210.
• Contributing guide, security policy and issue templates.

npm install piilite@0.2.1

piilite 0.2.0

First npm release of piilite: India-first PII detection and masking for JavaScript.

• Aadhaar (Verhoeff), PAN, GSTIN (mod-36), UPI, +91 phones, bank accounts, IFSC, voter ID, passport, DL, vehicle numbers, ABHA, UAN, ESIC, OTPs
• Indian names and addresses: Hinglish, 10 Indian languages, titles, S/O and W/O, PIN codes
• Presets: DPDP Act 2023, KYC, fintech, PCI, healthcare, HR, secrets, GDPR, HIPAA
• LLM helpers: fuzzy restore(), streaming redaction, redactObject()
• Express middleware, React hooks, CommonJS build, TypeScript types
• Optional piilite/openredaction bridge: 110 global ID types and document processors

Held-out benchmark: 84.3% recall, 99.7% precision (OpenRedaction 1.1.5: 45.7%, 76.7%).

npm install piilite