privatiser 0.5.0

Content anonymizer/pseudonymizer — redact sensitive data before sharing with AI

privatiser-engine

Open source anonymization engine powering Privatiser. Redacts IPs, API keys, secrets, PII, and cloud identifiers from any text - replacing them with structurally valid pseudonyms so context is preserved. Fully reversible.

Everything runs locally. Nothing leaves the machine.

Available as a Python library/CLI and a browser-native JavaScript port.

---

What it detects

Category	Examples	Pseudonym format
IP addresses	192.168.1.100, 10.0.0.0/16	10.x.x.x (preserves CIDR)
Email addresses	[e-mail]	[e-mail]
Domain names	prod-db.mycompany.com	redacted-host-1.example.net
MAC addresses	AA:BB:CC:DD:EE:FF	AA:BB:CC:00:00:01
AWS Account IDs	123456789012	100000000001
AWS ARNs	arn:aws:iam::123...:role/admin	Structure preserved, values redacted
S3 buckets	s3://my-prod-bucket	s3://redacted-bucket-1
API keys	AWS, OpenAI, Anthropic, Google, Groq, GitHub, Slack, Azure	REDACTED_SECRET_n
Connection strings	postgresql://user:pass@host/db	REDACTED_CONNSTR_n
JWT tokens	eyJhbG...	REDACTED_JWT_n
PEM private keys	-----BEGIN RSA PRIVATE KEY-----	REDACTED_PEM_KEY_n
Bearer tokens	Authorization: Bearer sk-...	REDACTED_BEARER_n
Generic secrets	password = "value"	Keyword preserved, value redacted
US phone numbers	[phone], [phone]	[phone]
UK phone numbers	[phone]	[phone]
Credit cards	4111 1111 1111 1111 (Luhn validated)	4000-0000-0000-0001
US SSN	123-45-6789	078-05-0001
Passports	C12345678	X00000001
IBAN	DE89370400440532013000	GB00XXXX000000000001
UUIDs	550e8400-e29b-41d4-...	00000000-0000-4000-a000-...
Azure / GCP IDs	Subscription IDs, project IDs	Redacted with counter

Skips well-known safe values: 127.0.0.1, 0.0.0.0, localhost, amazonaws.com, github.com, etc.

---

Python

Install

pip install privatiser

Usage

from privatiser import Privatiser

p = Privatiser()

text = 'server = "192.168.1.100"\npassword = "secret123"'
anonymized, mapping = p.anonymize(text)
# server = "10.0.1.8"
# password = "REDACTED_SECRET_1"

restored = p.deanonymize(anonymized, mapping)
assert restored == text  # perfect round-trip

Category toggles

p = Privatiser(enabled_categories={"pii": False})  # skip phone/card/SSN

Allowlist

p = Privatiser(allowlist=["localhost", "example.com"])  # never redact these

Custom patterns

from privatiser import Privatiser, register_custom

register_custom("ticket_id", r"TICKET-\d{4,6}", "REDACTED_TICKET_{n}")

p = Privatiser()
result, mapping = p.anonymize("Fix TICKET-12345")
# result: "Fix REDACTED_TICKET_1"

CLI

# From stdin
cat config.tf | privatiser anonymize

# From file, save mapping
privatiser anonymize config.tf -o clean.tf -m mapping.json

# Restore
privatiser deanonymize clean.tf -m mapping.json

# Disable categories
privatiser anonymize config.tf -d pii -d aws

---

JavaScript (browser / Node)

The privatiser.js file is a self-contained browser port with no dependencies. Drop it into any web project or use it in Node.

<script src="privatiser.js"></script>

const p = new Privatiser();
const { result, mapping } = p.anonymize(text);

// Restore
const restored = p.deanonymize(result, mapping);

Options

const p = new Privatiser({
  enabledCategories: { pii: false },          // disable a category
  allowlist: ["localhost", "example.com"],    // never redact these
  customWords: ["mycompany", "prod-server"],  // always redact these
});

---

How it works

1. Placeholder pass - before any pattern runs, detected values are replaced with null-byte markers (\x00PRIV_0\x00). This prevents patterns from matching inside already-redacted values.
2. Pattern priority - patterns run highest-priority first (connection strings before passwords, JWTs before base64, etc.).
3. Deterministic pseudonyms - the same value always gets the same pseudonym within a session, so repeated occurrences stay consistent.
4. Structural preservation - pseudonyms match the format of the original (IPs look like IPs, emails look like emails) so downstream tools and AI models aren't confused.
5. Restore pass - deanonymize() does a simple string replacement of pseudonyms back to originals using the mapping.

---

Project structure

src/privatiser/
  core.py          - Privatiser class, anonymize/deanonymize logic
  patterns/
    secrets.py     - API keys, JWTs, connection strings, PEM keys
    network.py     - IPs, domains, emails, MACs, URLs
    pii.py         - phone, credit card, SSN, passport, IBAN
    aws.py         - AWS account IDs, ARNs, S3 buckets
    cloud.py       - Azure, GCP identifiers
    identifiers.py - UUIDs, generic identifiers
  cli.py           - Click CLI entrypoint
  web/             - Flask web UI (optional)

privatiser.js      - Self-contained browser/Node JS port
tests/             - pytest test suite

---

Contributing

See CONTRIBUTING.md. Pattern contributions are especially welcome - if you work with a format that Privatiser doesn't detect yet, opening a PR with a new pattern + tests is the fastest way to get it added.

---

Attribution

MIT licensed - use it freely in personal and commercial projects. If you build something with it, a "Powered by Privatiser" credit is appreciated but not required.

License

MIT - see LICENSE.

Built and maintained by @XionDot. Web tool at privatiser.net.