🧽 scrubpii

Make any payload, log, or fixture safe to share — strip PII & secrets locally, in one pipe.

🌐 Try the browser playground →  ·  paste a payload, get a safe-to-share copy. Nothing is uploaded — it all runs client-side.

You need to attach a production API response to a GitHub issue. Or commit a realistic test fixture. Or drop a log into Slack so a teammate can repro. But it's full of emails, card numbers, JWTs, API keys, IPs, SSNs — and the one tool everyone reaches for, an online redactor, asks you to upload the very data you're trying to protect.

scrubpii finds and redacts PII & secrets from JSON, logs and text — 100% locally, deterministically, in one command. Nothing is uploaded. Nothing leaves your machine.

cat payload.json | npx scrubpii redact

- "email": "[email protected]",
+ "email": "[email protected]",
- "card_number": "4242 4242 4242 4242",
+ "card_number": "redacted_1",
- "authorization": "Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ...",
+ "authorization": "redacted_2",

Why scrubpii (and not a quick sed or an LLM)?

• 🔒 Local & deterministic. No server, no API key, no telemetry. Same input → same output, every time. Safe for secrets because it never sees the network.
• 🎯 Precision over noise. Credit cards are confirmed with the Luhn checksum, IBANs with mod-97, JWTs by decoding the header — so a random 16-digit order ID isn't flagged as a card. A redactor that cries wolf gets turned off.
• 🔗 Referential integrity. [email protected] becomes [email protected] everywhere it appears — so the redacted document still shows that two log lines are the same user. sed and online tools destroy that.
• 🧩 JSON-aware. Walks the document, redacts values whose key is sensitive (password, authorization, api_key — even with no detectable pattern) and scrubs PII hiding inside free-text fields. Output is still valid JSON, same shape.
• 🪶 Zero-dependency core. The library imports nothing at runtime and runs in the browser, Node, Deno and Bun. The CLI adds only cac + picocolors.

Why not just ask an LLM? You can't paste a 50 MB log into a chat, you can't put a chatbot in CI, and "redact the secrets" from a model is non-deterministic — it will miss some and hallucinate others. scrubpii is a checksum, not a guess.

Install

# zero-install, run it now
npx scrubpii scan payload.json

# or add it
npm install -g scrubpii      # global CLI
npm install -D scrubpii      # CI / pre-commit dependency
npm install scrubpii         # library

Node ≥ 18. Ships ESM + CJS + TypeScript types.

Quick start

# 1. See what's in there (read-only, nothing changes)
scrubpii scan payload.json

# 2. Get a safe copy on stdout
cat payload.json | scrubpii redact > safe.json

# 3. Use it as a CI / pre-commit guard — exit 1 if anything is found
scrubpii scan ./src --max-findings 0

scan — find, don't touch

$ scrubpii scan examples/app.log

examples/app.log (text)  10 findings
  1:42  Email address              a***@***.com
  2:41  JWT                        eyJ…90
  3:46  Credit card (Luhn-valid)   **** **** **** 4242
  4:45  URL credentials            svc…ss
  6:36  AWS access key             AKI…LE
  6:62  GitHub token               ghp…AB

Summary  4 secrets · 4 PII · 2 network across 1 file

Previews are always masked — scrubpii never prints the full secret, not even in its own reports.

redact — produce a safe copy

scrubpii redact users.json --out clean.json          # pseudonyms (default)
scrubpii redact users.json --strategy hash            # stable [EMAIL:32oeg5] tokens
scrubpii redact users.json --strategy label           # [EMAIL], [REDACTED]
scrubpii redact ./logs --in-place                      # rewrite a whole folder
echo "ping [email protected]" | scrubpii redact                  # pipe anything

| Strategy | [email protected] becomes | Keeps links? | Best for | | ------------ | ------------------- | :----------: | -------- | | pseudonym | [email protected] | ✅ | shareable, realistic fixtures (default) | | hash | [EMAIL:32oeg5] | ✅ | stable IDs across files, no counter state | | label | [EMAIL] | ❌ | quickest "just hide it" | | partial | a***@***.com | ❌ | keep a human hint | | remove | (deleted) | ❌ | strip entirely |

What it detects

Secrets — private keys, AWS access keys, GitHub / Slack / Stripe / OpenAI / Google / SendGrid keys, JWTs, user:pass@ URL credentials. PII — emails, Luhn-valid credit cards, mod-97 IBANs, US SSNs, phone numbers. Network — IPv4, IPv6, MAC addresses. JSON keys — any value under password, secret, token, authorization, api_key, cvv, pin, … (configurable), even when the value has no pattern.

Use it in your workflow

1. Sanitize before you paste. Bug report, Slack, screen-share — one pipe makes a prod payload safe to share, with relationships intact so it's still reproducible:

kubectl logs my-pod | scrubpii redact | pbcopy

2. Commit realistic fixtures from real data. Redact a prod export into test/fixtures with consistent pseudonyms, so foreign keys and "same user across rows" still hold — something hand-redaction always breaks:

scrubpii redact prod-export.json --strategy hash --out test/fixtures/users.json

3. Guard your repo / CI. Fail the build (or a pre-commit hook) the moment a real secret or customer record is about to be committed:

# .github/workflows/ci.yml
- run: npx scrubpii scan . --max-findings 0 --md scrubpii-report.md

# .husky/pre-commit
git diff --cached --name-only | xargs scrubpii scan --max-findings 0

Library API

import { redact, detect, Vault } from "scrubpii";

// Auto-detects JSON vs text:
const { output, findings, format } = redact(input, { strategy: "pseudonym" });

// Read-only detection with line/column (great for editors & CI):
for (const f of detect(logText)) {
  console.log(`${f.line}:${f.column}  ${f.label}  ${f.preview}`);
}

// Share one Vault to keep pseudonyms consistent across many documents:
const vault = new Vault("pseudonym");
const a = redact(file1, {}, vault);
const b = redact(file2, {}, vault); // same person → same alias in both

Also exported: redactText, redactJson, redactJsonValue, looksLikeJson, DETECTORS, luhnValid / ibanValid / jwtValid / ssnValid, and all types.

Configuration

scrubpii init writes scrubpii.config.json:

{
  "strategy": "pseudonym",
  "enable": [],            // [] = all types; or e.g. ["email", "credit-card"]
  "disable": ["phone"],    // turn specific detectors off
  "redactKeys": ["password", "token", "authorization", "..."],
  "salt": "",              // mixed into the hash strategy
  "maxFindings": 0          // CI gate threshold
}

CLI flags override the file: --only, --except, --strategy, --salt, --max-findings, --format json|text|auto, --config <file>.

Roadmap

• 🤖 Optional --ai layer (bring-your-own key). A pluggable NER pass to catch free-text names/addresses that rules can't — strictly opt-in; the core stays 100% offline and deterministic.
• 🔁 Reversible pseudonyms. Write the mapping to a local keyfile so a redacted fixture can be re-hydrated for debugging.
• ➕ More secret providers, custom regex detectors, and per-path ignore rules.
• 🌊 Streaming mode for multi-GB logs.
• ✅ 🌐 A browser playground (paste a payload, see redactions — nothing uploaded). Live here.

Issues and PRs welcome — especially new detectors with test vectors.

💖 Sponsor

scrubpii is free and MIT-licensed, built and maintained in spare time. If it saved you from leaking a customer record (or an afternoon of manual redaction), please consider supporting it:

• ⭐ Star this repo — the simplest free way to help others find it.
• 🍋 Sponsor via Lemon Squeezy — one-time or recurring.

License

MIT © scrubpii contributors