Security Architecture — Zero-Knowledge Offline Anonymization

Design Principles

anonym.plus is built on three security principles:

• Offline-first: The entire PII detection and anonymization pipeline runs locally. Documents are never uploaded to any server.
• Zero-knowledge: Your password never leaves your device. Only cryptographic proofs are sent during authentication. The server cannot access your vault contents.
• Defense in depth: Encryption at rest (AES-256-GCM), key derivation via Argon2id, automatic vault locking, and a 24-word BIP39 recovery phrase provide layered protection.

Threat Model

The following table describes what anonym.plus protects against and what is outside the scope of its security guarantees.

Threat	Mitigation	Status
Documents sent to external servers	All processing runs locally via bundled Presidio + spaCy. No network calls for PII detection or anonymization.	Mitigated
Vault data accessed by attacker with disk access	Vault encrypted with AES-256-GCM. Key derived from user password via Argon2id (64 MB memory, 3 iterations). Without PIN or recovery phrase, data is unreadable.	Mitigated
Server compromise exposes user passwords	Zero-knowledge authentication. Passwords are hashed client-side before any server communication. Server stores only cryptographic proofs.	Mitigated
Encryption keys extracted from memory	Key material is zeroed from memory when the vault locks. Auto-lock triggers after 15 minutes of inactivity.	Mitigated
Man-in-the-middle on API communication	All API communication over HTTPS/TLS. Certificate pinning not implemented (standard browser/Tauri TLS).	Partially mitigated
Malware on user's device	Outside scope. If an attacker has code execution on your machine, no desktop app can guarantee security. Use OS-level protections.	Out of scope
NLP detection misses PII entities	Presidio + spaCy use both NER (machine learning) and regex patterns. Configurable confidence thresholds and manual review step. False negatives are possible — always review results.	Partially mitigated

Encryption Architecture

Vault Encryption

All local data (processing history, encryption keys, settings) is stored in an encrypted vault.

Algorithm

AES-256-GCM (authenticated encryption with associated data)

Key Derivation

Argon2id with 64 MB memory cost, 3 iterations, 32-byte output

Recovery Phrase

24-word BIP39 mnemonic (256 bits of entropy), generated during first setup

Quick Unlock

Optional 4-8 digit PIN for convenience. After 3 failed attempts, recovery phrase required.

Auto-lock

Vault locks after 15 minutes of inactivity (no mouse, keyboard, or scroll events)

Key Zeroing

Encryption key material is zeroed from memory when the vault locks

Document Encryption (Reversible Anonymization)

When using the "Encrypt" anonymization operator, PII entities are encrypted with per-key AES-256-GCM so they can be decrypted later via the Deanonymize feature.

Algorithm

AES-256-GCM with random nonce per entity

Key Lengths

128-bit, 192-bit, or 256-bit (256-bit recommended and default)

Key Storage

Zero-knowledge: key values never leave the Rust backend. Frontend only references keys by ID.

Key Rotation

Generates new key material. Old key value is permanently replaced. Documents encrypted with old key require old key value for decryption.

Authentication

Password Handling

Passwords are hashed client-side using Argon2id + HKDF + SHA-256 before any server communication. The server never receives plaintext passwords.

Session Tokens

JWT with separate secrets for user and admin authentication. Tokens stored in sessionStorage (cleared on tab close).

Admin Authentication

Separate JWT secret (ADMIN_JWT_SECRET). 12-character minimum password. TOTP-based two-factor authentication.

What the Server Sees

The anonym.plus server handles account management, licensing, and license purchases. Here is exactly what it can and cannot access:

• Can access: Email address, license status, machine activations, support tickets, payment metadata (via Stripe/PayPal).
• Cannot access: Your documents, PII detection results, anonymized outputs, encryption keys, vault contents, processing history, or password.

For a full breakdown of collected data, see the Privacy Policy.

Responsible Disclosure

Independent Review — Current Status

Everything above describes how anonym.plus is built. It is a description by the people who built it, and it is worth being explicit about what that is and is not:

• anonym.plus has not undergone an independent security audit or penetration test. When one is commissioned, this section will name the auditor, the date, the scope, and what was found.
• anonym.plus holds no security certification of its own. Where this site mentions ISO 27001, it refers to the certification held by the operator of the data centre that hosts our servers — not to us.
• The claims on this page are written to be checkable rather than taken on trust: the absence of network traffic during anonymisation, the code signature on the installer, and the hash verification of downloaded language packs can all be verified without our cooperation.

Open Source Components

anonym.plus uses the following open-source components for its security-critical functions:

• Microsoft Presidio — PII detection engine (Apache 2.0)
• spaCy — NLP models for named entity recognition (MIT)
• Tesseract OCR — Image text extraction (Apache 2.0)
• Tauri — Desktop app framework with Rust backend (MIT/Apache 2.0)
• Argon2 — Password hashing / key derivation (CC0)

Related Documents

• Privacy Policy — what we store about your account, and who else is involved in operating the service
• Terms of Service — including our role under the GDPR with respect to your documents and your account data

References

• Microsoft Presidio — GitHub repository
• spaCy — Industrial-strength NLP
• RFC 5116 — AES-GCM authenticated encryption specification
• RFC 9106 — Argon2 memory-hard function specification
• BIP39 — Mnemonic code specification
• GDPR — General Data Protection Regulation full text