Skip to main content
Security Model
Threat Model
Trusted Zone
Your Browser
Crypto
Encryption
Hashing
Key Derivation
User Data
Data stays local
Input: Text, Passwords, Files
Passwords
Files
External Zone
Server
No access to what you type, paste, or upload
Network
Your input is not transmitted
Third Parties
No data sharing
Security Assumptions
- • Your device is secure
- • Modern browser with Web Crypto API
- • Strong passwords
Limitations
- • Compromised device = risk
- • Browser extensions with broad permissions may access page content
- • We cannot recover lost passwords - encryption is mathematically irreversible
Security properties and explicit limitations of our client-side architecture.
Trust Assumptions
- Device free from malware
- Standard, up-to-date browser
- No compromised extensions
- Secure network (HTTPS)
- Understanding of crypto limitations
Explicit Limitations
- Does not protect against endpoint threats like keyloggers, clipboard monitors, or screen recording software.
- Cannot prevent you from accidentally sharing sensitive output (e.g., copying an encrypted token to an untrusted chat application).
- Assumes correct use of cryptographic parameters. The security of AES-GCM encryption, for example, relies on the user providing a strong, unique password.
- Does not provide authentication - anyone with the password can decrypt the data.
- Does not protect against replay attacks if encrypted data is captured and re-submitted.
Key lifecycle and container format
This section documents the parameters the tool actually uses. They come from ENCRYPT_CONFIG in the repository, so a change there moves the interface, the worker and the container header together.
- Key derivation: PBKDF2-HMAC-SHA-256, 100,000 iterations, 32-byte key. A fresh 16-byte salt is generated for every encryption with crypto.getRandomValues().
- Encryption: AES-256-GCM with a 12-byte IV and a 16-byte authentication tag. The derived key stays inside the browser worker and is never written to storage.
- Text output: base64 or hex of salt, IV and ciphertext. Salt and IV are also shown separately, so decryption does not have to guess the parameters.
- File output: a 40-byte ZVE1 header (magic, version, iteration count, salt, 1 MiB chunk size, chunk count, total size), then for each chunk a 12-byte IV and the ciphertext with its own tag.
- The chunk index is bound as AAD, so reordered or swapped chunks are rejected by GCM instead of producing wrong plaintext.
Compatibility and limits
- Files written before this container existed (salt | IV | ciphertext+tag, no magic) are still decrypted automatically.
- Limits: 50 MB per file, 16 MiB per chunk, 1,048,576 chunks. Larger input is rejected before any crypto call.
- Nothing is persisted: closing the tab discards the password-derived key, and only the service-worker cache of the app shell remains (see the storage disclosure).