Zero-Knowledge Password Manager, Explained
“Zero-knowledge” is the most abused phrase in the password manager market, so this page makes no appeal to trust. Below is what VaultKeepR actually does, primitive by primitive, with the source you can read. The claim is narrow: the service that stores your vault has no technical ability to read it. Everything else, what that guarantees, what it forbids, and what it cannot save you from, follows from the code.
What Zero Knowledge Means, Precisely
A zero knowledge design answers one question with no: could the provider read your vault if it wanted to, or was legally compelled to? If encryption keys exist only on your devices, if encryption happens before any upload, and if no recovery path can reconstruct those keys, the answer is no. The servers hold noise. A subpoena produces noise. A rogue employee gets noise.
Most services that call themselves zero knowledge stop at the first condition and quietly keep an email reset that undoes the other two. The test is always the recovery design, because that is where the property gets traded away.
The Proof Is in the Code
The cryptography core is public under the MIT license: github.com/VaultKeepR/vaultkeepr-public. These are the load-bearing pieces:
- Vault encryption: XChaCha20-Poly1305, authenticated encryption with a random 192-bit nonce per operation. Every entry is sealed on your device before storage.
- Key derivation: Argon2id with time cost 3, memory cost 64 MiB, parallelism 4. Memory-hard derivation is what makes offline brute-force on a stolen vault expensive.
- Session keys: the SessionKeyStore keeps key material in memory for the duration of an unlock, derived again on the next unlock. Keys are not written to disk alongside the vault.
- Integrity commitments: HMAC-SHA256 binds each vault version to its key material, so a tampered or substituted blob fails verification instead of decrypting.
Read it, build it, run it. The parameters above are the ones in the repository, not aspirations, and any divergence between this page and the code is a bug report.
What Zero Knowledge Guarantees
The server, in every storage mode, only ever sees:
- Ciphertext: sealed vault blobs it can store and replicate but not open.
- A content address: where to find the latest version.
- Wallet signatures: proof you control the account, which reveals nothing about the vault's contents.
From that follows breach resilience by construction. A compromise of the infrastructure yields encrypted blobs that cost an attacker an offline brute-force against Argon2id, per vault, with no shortcut through the vendor.
What It Forbids, on Purpose
Zero knowledge has a price, and it is worth stating it plainly: there is no reset by email. Not as a policy choice that could be reversed, but as a consequence of the architecture. An email reset that hands back vault access would require the key, which would require the key to exist server-side, which would end the property. Support cannot recover your master secret. Nobody can.
Products that offer both “zero knowledge” and “lost password recovery” are offering one of them in name only. The recovery question is where every zero knowledge claim lives or dies.
What It Cannot Save You From
Honest accounting includes the failure mode. If you lose your master key, you lose the vault. No support ticket, no identity check, no court order recovers it, because nothing on the server side can decrypt it. This is the same property read from the other direction.
That is why recovery is designed around fragments instead of escrow. An optional Shamir 3-of-5 scheme splits a recovery secret into five independent fragments, kept in separate places or with separate people, and any three rebuild access. Two fragments are useless, so losing one location does not lock you out or leak anything. The scheme keeps recovery humanly possible without ever creating a master backdoor.
- Lose your key with no fragments: the vault stays sealed, permanently.
- Lose one or two fragments: three remaining fragments still recover access.
- An attacker with two fragments: still locked out, by design.
Where to Go From Here
The broader landscape, cloud versus self-hosted versus decentralized, is covered in the password manager guide. Where the ciphertext lives, including running your own IPFS node, is on the self-hosted page. And since every claim here reduces to “read the code”, the shortest verification path is to download VaultKeepR and keep vaultkeepr-public open in a tab while you do it.