metamorphic-crypto
Zero-knowledge end-to-end encryption library with post-quantum hybrid KEM, hybrid PQ signatures, and an opt-in CNSA 2.0 suite axis (matched-strength hybrid + pure ML-KEM-1024 / ML-DSA-87 / AES-256-GCM).
Built for Metamorphic and Mosslet — privacy-first apps by Moss Piglet Corporation where all user data is encrypted client-side and the server only stores opaque ciphertext.
What this provides
- Secretbox (XSalsa20-Poly1305) — symmetric authenticated encryption
- Sealed box (X25519) — anonymous public-key encryption (libsodium-compatible)
- Hybrid PQ KEM (ML-KEM-512 + X25519) — NIST Cat-1 post-quantum key encapsulation (opt-in)
- Hybrid PQ KEM (ML-KEM-768 + X25519) — NIST Cat-3 post-quantum key encapsulation (default)
- Hybrid PQ KEM (ML-KEM-1024 + X25519) — NIST Cat-5 post-quantum key encapsulation (opt-in)
- Argon2id KDF — password-based key derivation (libsodium INTERACTIVE parameters)
- Hybrid PQ signatures (ML-DSA + Ed25519) — NIST Cat-2/3/5 composite digital signatures (strict AND)
- CNSA 2.0 suite axis (opt-in) — matched-strength hybrid (X448 / P-521 / Ed448 / ECDSA-P-521) and pure post-quantum (ML-KEM-1024, ML-DSA-87, AES-256-GCM)
- Hashing (SHA3-512/256, SHA-256/512) — public, one-shot digest functions (e.g. for key fingerprints / safety numbers)
- Verifiable random function (ECVRF-EDWARDS25519-SHA512-TAI, RFC 9381) — classical VRF for transparency-log index privacy (CONIKS-style)
- WASM bindings — browser-ready via
wasm-pack - Recovery keys — human-readable base32 encoding for key backup
Security levels
| Level | ML-KEM | NIST Category | Equivalent | Version Tag | Default |
|---|---|---|---|---|---|
| Cat-1 | 512 | 1 | ~AES-128 | 0x01 |
No |
| Cat-3 | 768 | 3 | ~AES-192 | 0x02 |
Yes |
| Cat-5 | 1024 | 5 | ~AES-256 | 0x03 |
No |
NIST (FIPS 203) standardizes ML-KEM only at categories 1/3/5 — there is no category-2/4 parameter set, so none is offered. All levels use the same combiner construction. The classical half is X25519 (~Cat-1 classical) at every tier — it does not scale up with the ML-KEM parameter set; at Cat-3/Cat-5 the post-quantum half dominates and X25519 is the classical floor (standard hybrid-KEM practice: a break requires defeating both halves). hybrid_open auto-detects the level from the version tag byte — old and new ciphertext coexist seamlessly.
Security properties
#![forbid(unsafe_code)]— no unsafe anywhere in the crate- All secret key material zeroized after use
- Constant-time MAC comparison via RustCrypto
- OS CSPRNG via
getrandom(no userspace PRNG) - Hybrid construction: both ML-KEM AND X25519 must be broken to compromise a sealed key
Hybrid KEM construction
The hybrid combiner matches the format used by @noble/post-quantum's ml_kem768_x25519:
Seed expansion: SHAKE256(seed_32) → 96 bytes [ML-KEM seed (64) || X25519 sk (32)]
Combiner: SHA3-256(ss_mlkem || ss_x25519 || ct_x25519 || pk_x25519 || label)
Cat-1 (ML-KEM-512, opt-in)
Public key: ML-KEM-512 ek (800 B) || X25519 pk (32 B) = 832 bytes
Ciphertext: 0x01 || ML-KEM-512 ct (768 B) || X25519 eph pk (32 B) || nonce (24 B) || secretbox ct
Cat-3 (ML-KEM-768, default)
Public key: ML-KEM-768 ek (1184 B) || X25519 pk (32 B) = 1216 bytes
Ciphertext: 0x02 || ML-KEM-768 ct (1088 B) || X25519 eph pk (32 B) || nonce (24 B) || secretbox ct
Cat-5 (ML-KEM-1024, opt-in)
Public key: ML-KEM-1024 ek (1568 B) || X25519 pk (32 B) = 1600 bytes
Ciphertext: 0x03 || ML-KEM-1024 ct (1568 B) || X25519 eph pk (32 B) || nonce (24 B) || secretbox ct
Targets
| Target | Build | Use case |
|---|---|---|
| Native | cargo build |
Tests, CLI tools, Elixir NIF (metamorphic_crypto Hex package) |
| WASM | wasm-pack build --target web |
Browser (Phoenix LiveView, any SPA) |
| iOS | UniFFI (planned) | Native Swift apps |
| Android | UniFFI (planned) | Native Kotlin apps |
Usage
use ;
use ;
use ;
use ;
// Symmetric encryption
let key = generate_key;
let ciphertext = encrypt_secretbox_string.unwrap;
let plaintext = decrypt_secretbox_to_string.unwrap;
assert_eq!;
// Hybrid PQ seal (Cat-3, default)
let kp = generate_hybrid_keypair;
let sealed = hybrid_seal.unwrap;
let opened = hybrid_open.unwrap;
// Hybrid PQ seal (Cat-5)
let kp5 = generate_hybrid_keypair_1024;
let sealed5 = hybrid_seal_1024.unwrap;
let opened5 = hybrid_open.unwrap; // auto-detects level
// Hybrid PQ seal (Cat-1)
let kp1 = generate_hybrid_keypair_512;
let sealed1 = hybrid_seal_512.unwrap;
let opened1 = hybrid_open.unwrap; // auto-detects level
Hashing
Public, one-shot digest functions over the already-present, audited sha3 and
sha2 dependencies. These are intended for public data only — key
fingerprints / safety numbers and key-transparency-log entries — where both the
input (e.g. a public key) and the output digest are meant to be public.
sha3_512 is the recommended default (NIST Cat-5, ~256-bit collision
resistance, consistent with the crate's Keccak-based combiner). sha3_256,
sha256, and sha512 are provided so integrators can match an existing format.
use ;
// Take raw bytes, return fixed-size byte arrays.
let digest: = sha3_512; // recommended default
let d256: = sha3_256;
let s256: = sha256; // SHA-2 interop
let s512: = sha512; // SHA-2 interop
// Encode the digest yourself when needed:
use b64;
let fingerprint_b64 = encode;
Domain separation (recommended for fingerprints / transparency logs)
For key fingerprints, safety numbers, and key-transparency-log entries, prefer
sha3_512_with_context, which binds the digest to a versioned context label so
the same bytes hashed for different purposes can never collide or be
reinterpreted across contexts. It is exactly as strong as sha3_512 — it is
SHA3-512, over an unambiguously framed message — and makes intent explicit:
use sha3_512_with_context;
let fp = sha3_512_with_context;
let log = sha3_512_with_context;
// fp and log are unrelated even if the byte inputs coincide.
Stable wire format (reproduce exactly for cross-language parity):
SHA3-512( u64_be(len(context_utf8)) || context_utf8 || data )
The 8-byte big-endian length prefix makes the (context, data) boundary
unambiguous (no boundary-confusion collisions). Use a versioned namespace label.
Encoding: the native functions take &[u8] and return raw byte arrays — encode
to base64 or hex at the call site. The WASM bindings take/return base64 to match
the rest of the WASM API (see below).
Do not hash secrets with these. A bare hash makes no guarantees about its
inputs, and (consistent with the rest of the crate) the hashing path adds no
zeroize/constant-time ceremony — wiping a transient copy of already-public data
would add cost without protection. If you need to process secret material
(passwords, private keys), use the right construction instead — this crate's
Argon2id derive_session_key for password-based derivation, or a dedicated
KDF/MAC. The encryption APIs that handle secrets already zeroize on drop.
Verifiable random function (ECVRF, RFC 9381)
The vrf module exposes ECVRF-EDWARDS25519-SHA512-TAI — RFC 9381 ciphersuite
0x03: Edwards25519 + SHA-512 + the try-and-increment hash-to-curve. A VRF lets
the key owner compute, for any input alpha, a pseudorandom output beta plus a
proof pi that beta is correct under their public key. Anyone with the public
key can verify pi, but cannot compute beta for a new input and cannot learn
alpha from (pi, beta). This is the primitive behind transparency-log index
privacy (CONIKS-style): a directory maps a private identity index to a
verifiable, pseudorandom tree position without revealing which identities it
holds.
It is built on the same curve25519-dalek backend as the Ed25519 interop
module — no new curve stack — and is pinned byte-for-byte by RFC 9381's own test
vectors.
use ;
let = ecvrf_generate_keypair;
let alpha = b"identity index";
let pi = ecvrf_prove?; // 80-byte proof
let beta = ecvrf_verify?; // Ok(Some(beta)) if valid
assert!;
| Item | Bytes | Layout |
|---|---|---|
| secret key | 32 | Ed25519-style seed |
| public key | 32 | compressed Edwards Y = x*B |
proof pi |
80 | Gamma(32) || c(16) || s(32) |
output beta |
64 | SHA-512(0x03 || 0x03 || cofactor*Gamma || 0x00) |
Honest posture. This VRF is classical (elliptic-curve discrete log). It
protects exactly one property — index privacy — and is the one non-post-quantum
piece in the transparency stack; integrity, authenticity, confidentiality, and
hash-based commitments are post-quantum independently of it. RFC 9381's sibling
ECVRF-EDWARDS25519-SHA512-ELL2 (0x04, constant-time Elligator2 hash-to-curve)
is a designed-in future addition that lands when the released curve backend
exposes a conformant hash-to-curve (curve25519-dalek 5.x); because the suite
octet is bound into every hash, adding it is purely additive and never
invalidates a 0x03 proof. A hybrid (post-quantum + classical) VRF is intended
for when an audited lattice VRF exists; none does today, so it is not built.
These primitives are not FIPS-validated.
Hybrid PQ signatures
Composite digital signatures: every message is signed by both ML-DSA (FIPS 204) and Ed25519 (RFC 8032), and verification requires both to be valid (strict AND). An attacker has to break both a lattice scheme and an elliptic-curve scheme to forge, and cannot strip one algorithm to downgrade the other. This is the signing counterpart to the hybrid KEM above.
use ;
let kp = generate_signing_keypair; // Cat-3 (ML-DSA-65 + Ed25519), default
let sig = sign.unwrap;
assert!;
// Re-derive the public key from a backed-up secret key:
use derive_public_key;
assert_eq!;
Cat-2 (generate_signing_keypair_44) and Cat-5 (generate_signing_keypair_87)
are also available; verify auto-detects the level from the signature's version
tag. The secret_key field is zeroized on drop.
Signing levels and mode
| Level | ML-DSA | NIST Category | Equivalent | Version Tag | Default |
|---|---|---|---|---|---|
| Cat-2 | ML-DSA-44 | 2 | ~AES-128 | 0x01 |
No |
| Cat-3 | ML-DSA-65 | 3 | ~AES-192 | 0x02 |
Yes |
| Cat-5 | ML-DSA-87 | 5 | ~AES-256 | 0x03 |
No |
ML-DSA is signed with the hedged (randomized) variant — FIPS 204's default and most conservative mode (resilient to RNG failure, hardened against fault / side-channel attacks that deterministic lattice signing invites). Ed25519 is deterministic per RFC 8032. As a result signature bytes are non-reproducible, but the wire format is deterministic and pinned.
Domain separation and wire format
Both algorithms sign the same domain-separated message, framed exactly like
sha3_512_with_context (a length-prefixed context):
signed_msg = I2OSP(len(context_utf8), 8) || context_utf8 || message
ML-DSA signs signed_msg with an empty native context, so the framing is
identical for both algorithms and across every language binding. Byte layout
(Ed25519 first, fixed-size, so the ML-DSA tail needs no length prefix):
signature = tag || ed25519_sig (64 B) || ml_dsa_sig (2420 / 3309 / 4627 B)
public_key = tag || ed25519_pk (32 B) || ml_dsa_pk (1312 / 1952 / 2592 B)
secret_key = tag || ed25519_seed(32 B) || ml_dsa_seed(32 B) = 65 B
Dependency audit posture
| Dependency | Version | Audited | Notes |
|---|---|---|---|
ed25519-dalek |
2.x | Yes (mature) | Widely deployed RFC 8032 implementation. |
ml-dsa |
0.1.x | No (RustCrypto) | FIPS 204 (final). New crate, not yet independently audited. Pinned; tracked for the FIPS-mode roadmap. |
ML-DSA is defense-in-depth on top of the independently-strong Ed25519: even if a
flaw were found in the young ml-dsa implementation, the composite remains at
least as strong as Ed25519. This is stated honestly so integrators can choose
while the post-quantum implementation matures toward audit / FIPS validation.
CNSA 2.0 suite axis (opt-in)
By default everything above is Suite::Hybrid — the classical+PQ strict-AND
constructions (ML-KEM + X25519; ML-DSA + Ed25519). If you have no specific
mandate, that is the recommended choice and you can ignore this section.
For deployments that must follow the NSA's Commercial National Security
Algorithm Suite 2.0 (CNSA 2.0 / NIST IR 8547), a Suite axis lets you raise
the posture with a single extra argument. It is orthogonal to the
SecurityLevel (Cat-1/3/5) you already know, so you really have two independent
knobs:
posture (Suite) × parameter set (SecurityLevel)
┌──────────────────────────────┐ ┌───────────────────────┐
│ Hybrid (default) │ │ Cat-1 / Cat-3 / Cat-5 │
│ HybridMatched (opt-in) │ └───────────────────────┘
│ PureCnsa2 (opt-in) │
└──────────────────────────────┘
| Suite | What it is | Classical partner | Status |
|---|---|---|---|
Hybrid |
Existing strict-AND classical+PQ. Byte-for-byte unchanged. | X25519 / Ed25519 (every tier) | Default, recommended |
HybridMatched |
Classical partner matched to the PQ category so it is never the weak link | KEM: Cat-3→X448, Cat-5→P-521 ECDH · Sign: Cat-3→Ed448, Cat-5→ECDSA-P-521 | Opt-in |
PureCnsa2 |
Pure post-quantum, no classical half (the CNSA-2.0 box) | none | Opt-in, Cat-5 only |
HybridMatched at the lowest rung (KEM Cat-1 / sign Cat-2) is identical to
Hybrid — no new format is produced there, so nothing breaks.
New wire formats (new suites only)
The Hybrid suite (and HybridMatched at the lowest rung) keep their existing
0x01/0x02/0x03 ciphertext tags and byte layout untouched. The matched / pure
suites use new tags and a CNSA-correct seal envelope:
| Suite + level | KEM | Tag |
|---|---|---|
PureCnsa2 Cat-5 |
ML-KEM-1024 + AES-256-GCM | 0x10 |
HybridMatched Cat-3 |
ML-KEM-768 + X448 + AES-256-GCM | 0x13 |
HybridMatched Cat-5 |
ML-KEM-1024 + P-521 ECDH + AES-256-GCM | 0x14 |
ikm = ss_mlkem (PureCnsa2) | ss_mlkem || ss_ecc (HybridMatched)
key = HKDF-SHA512(ikm, info = suite_tag || context_label) -> 32-byte AES-256 key
out = AES-256-GCM(key, 96-bit random nonce, AAD = suite_tag || context_label)
wire = tag(1) || kem_ct || [ecc_eph_pk] || nonce(12) || ct || gcm_tag(16)
Each encapsulation yields a fresh KEM secret, so the derived AES-256 key is
single-use and the random 96-bit nonce can never repeat — SIV-grade misuse
resistance without leaving the CNSA-approved set (no AES-GCM-SIV). Note the
deliberate hash split: HKDF-SHA512 for key derivation here; SHA3-512
stays the choice for leaf/transcript hashing (sha3_512_with_context).
Context labels
The new suites bind a versioned context label into both the HKDF info and the
GCM AAD (and, for signatures, the I2OSP-framed message). Grammar:
"<namespace>/<purpose>/v<major>". The namespace is the one per-tenant knob;
the protocol shape stays fixed. Library defaults are SEAL_CONTEXT_V1
("metamorphic/seal/v1") and SIGN_CONTEXT_V1 ("metamorphic/sign/v1"); pass
your own (e.g. "mosslet/seal/v1") to namespace your deployment.
Usage (Rust)
use ;
// --- KEM / seal: the pure CNSA-2.0 box (ML-KEM-1024 + AES-256-GCM) ---
let kp = generate_hybrid_keypair_suite.unwrap;
let sealed = hybrid_seal_suite.unwrap;
// `hybrid_open` auto-detects the tag using the DEFAULT context label; if you
// sealed with a custom label, open with it explicitly:
let opened = hybrid_open_with_context.unwrap;
// --- Signatures: ML-DSA-87 only (Cat-5 pure) ---
let sk = generate_signing_keypair_suite.unwrap;
let sig = sign.unwrap;
assert!;
// `sign` / `verify` / `derive_public_key` auto-detect the suite from the version
// tag — no suite argument is needed once the key exists.
seal_for_user_with_suite is the user-facing seal that falls back to legacy
X25519 when no PQ key is present, mirroring seal_for_user_with_level.
Honest claims
Claim: "CNSA 2.0 algorithm suite, NCC-audited components, pure-Rust,
memory-safe (forbid-unsafe)." Not "FIPS 140-3 validated." PureCnsa2 is
more standards-compliant but leans entirely on the (not-yet-independently-audited
at our layer) lattice implementation, which is exactly why the strict-AND
Hybrid default stays recommended: it keeps the classical backstop until the PQ
implementations are audited / validated.
WASM (browser)
import init from './pkg/metamorphic_crypto.js';
await ;
const key = ;
const ciphertext = ;
Hashing (WASM)
Digest exports take base64-encoded input and return the digest as base64. Decode or re-encode to hex on the JS side if a hex fingerprint is required.
import init from './pkg/metamorphic_crypto.js';
await ;
const dataB64 = ;
const digestB64 = ; // also: sha3_256, sha256, sha512
// Domain-separated (recommended for fingerprints / transparency logs):
const fp = ;
Signatures (WASM)
Keys and signatures are base64; the message is base64 and context is a UTF-8
string. verify returns true only if both component signatures are valid.
import init from './pkg/metamorphic_crypto.js';
await ;
const kp = ; // { publicKey, secretKey }
const msg = ;
const sig = ;
const ok = ; // true
CNSA 2.0 suites (WASM)
The Suite axis is exposed as a string argument ("hybrid" (default),
"hybridMatched", or "pureCnsa2") alongside the usual "cat1"/"cat3"/"cat5"
level. Decryption / verification auto-detect the suite from the version tag.
import init from './pkg/metamorphic_crypto.js';
await ;
// Pure CNSA-2.0 KEM box (ML-KEM-1024 + AES-256-GCM)
const kp = ; // { publicKey, secretKey }
const sealed = ;
// Open with the context label used at seal time (default "metamorphic/seal/v1"):
const opened = ; // base64
// Pure ML-DSA-87 signatures
const sk = ;
const sig = ;
const ok = ; // true
For a custom per-tenant namespace, use hybridSealSuiteWithContext(..., "mosslet/seal/v1") and open with the same label. sealForUserWithSuite mirrors
sealForUser with the suite/level appended.
Tests
License
Dual-licensed under MIT or Apache-2.0 at your option.