Free, open-source, browser-native fully homomorphic encryption by jesed.
⚠️ Pre-audit (v0.1). Verified math, pending third-party audit before v1.0.
RingCrypt lets you encrypt data, compute on it, and decrypt only the result — sums, products, dot products, averages — all on ciphertext. The host running the computation never sees your plaintext.
| Capability | Status |
|---|---|
| NTT / INTT core (O(N log N) polynomial multiply) | ✅ |
| CKKS scheme: encode, encrypt, decrypt, add, multiply | ✅ |
| Threshold secret sharing (private mean) | ✅ |
| Browser live demo (WASM) | ✅ |
| GPU NTT compute shaders (wgpu/Vulkan), verified bit-exact on RTX 3060 | ✅ |
| Modulus chain / rescaling / relinearization | 🚧 |
Download from releases or build from source:
git clone git@github.com:jesedv/ringcrypt.git &&cd ringcrypt
# Full FHE workflow — encrypt, compute, decrypt
cargo run --release -- keygen --out keys/
echo'[120, 95, 132, 88, 110]'> data.json
cargo run --release -- encrypt --pub keys/pub.json --in data.json --out ct.json
cargo run --release -- decrypt --sec keys/sec.json --in ct.json
# Self-tests + benchmarks (no args)
cargo run --release
cargo test --workspace # 601 self-test checks
cargo run --release --bin gpu-bench # GPU NTT benchmarkringcrypt Run self-tests + benchmarks
ringcrypt keygen --out <dir> Generate public + secret key
ringcrypt encrypt --pub <pk> --in <data> --out <ct>
ringcrypt compute add <a> <b> --out <r>
ringcrypt compute mul <a> <b> --out <r>
ringcrypt compute sum <a> <b> [c...] --out <r>
ringcrypt decrypt --sec <sk> --in <ct>
Input files: JSON array [1.0, 2.0, 3.0] or plaintext (one number per line).
N=256 GPU: 1741 µs CPU: 7 µs PASS
N=512 GPU: 1783 µs CPU: 17 µs PASS
N=1024 GPU: 1814 µs CPU: 34 µs PASS
N=2048 GPU: 1840 µs CPU: 81 µs PASS
N=4096 GPU: 1891 µs CPU: 178 µs PASS
All bit-exact with CPU reference. GPU overhead dominates at small N — wins at larger sizes.
RingCrypt lives in the polynomial ring R = Z_q[x]/(x^N+1). Polynomial multiplication — the dominant cost of every FHE operation — is accelerated by the Number-Theoretic Transform (finite-field FFT), turning O(N²) into O(N log N).
ringcrypt/
├── crates/
│ ├── ringcrypt-ntt/ # NTT/INTT, RLWE negacyclic multiply, Barrett modmul
│ ├── ringcrypt-scheme/ # CKKS: canonical embedding, encrypt/decrypt, homomorphic ops
│ ├── ringcrypt-ss/ # Threshold secret sharing (p = 2^31 − 1)
│ └── ringcrypt-wasm/ # wasm-bindgen bridge
├── examples/ # encrypted workflow, average, dot product
├── web/ # Svelte static site + live WASM demo
├── scripts/ # build, test, bench
└── docs/ # Math exposition, publishing guide
| Parameter | Value |
|---|---|
| Polynomial degree N | 128 (64 complex slots) |
| Ciphertext modulus Q | 2⁶⁴ − 2³² + 1 (Goldilocks/Solinas) |
| Scale Δ | 2²⁴ (~7 decimal digits) |
| Secret key | Ternary (−1, 0, 1) |
| Noise σ | 3.2 |
The real engine runs in your browser — zero servers, zero trust:
- NTT self-test — verifies transforms, convolutions, and modular arithmetic
- CKKS self-test — encode/decode roundtrip, encrypt/decrypt, homomorphic add + multiply
- Private mean — five parties reveal no values, all learn the mean
- No unsafe in core.
#![forbid(unsafe_code)]across the NTT and scheme crates - Correctness over speed. Every kernel is cross-checked against a reference
- Reproducible. Seeded PRNGs; the self-test runs identically everywhere
- Browser-native. Same Rust → native CLI + WASM
FHE lives in polynomial rings — Z_q[x]/(x^N+1). The ring is the cryptosystem. The name is the math.
MIT © 2026 RingCrypt contributors.