Threat model

What we protect, and what we don't

A clear boundary matters more than a list of promises.

Defended by construction

  • Host operator reading stored fragments
  • Theft or imaging of one host machine
  • Full compromise of our database or storage
  • A host returning altered shard bytes
  • Loss of several shards per file
  • Insider access to customer files

Out of scope

  • Malware or keyloggers on your device
  • Weak or reused passphrases
  • Lost passphrases — there is no reset
  • Metadata needed to route and bill
  • Hostile browser extensions in your vault
  • Network-wide traffic analysis

Primitives in use

Key derivationPBKDF2-HMAC-SHA256, 600,000 rounds, random salt
File encryptionAES-256-GCM with a unique key and IV per file
Key wrappingContent keys sealed under the master key
FragmentationReed-Solomon erasure coding, k-of-n recovery
IntegritySHA-256 per shard and manifest digest
TransportTLS 1.3; payloads are ciphertext
AuthorizationRow-level security scoped to owner or host

Metadata we necessarily hold

Zero-knowledge applies to content, not existence. We keep account identity, file names, sizes, shard counts, digests, node assignments, and timestamps to bill, route, and repair. Hosts see only a shard ID and byte length.