APIsPricingDocsWhite PaperTokenBlogAboutSecurity Demo
Log InGet API Key
H33-VaultH33-ShareH33-ShieldH33-HealthH33-KeyH33-128H33-CKKSH33-256H33-FHE-IQH33-TFHEFHE OverviewH33-CompileZK LookupsBiometricsH33-3-KeyH33-MPCZK-TrustlessZK-PhishZK-VerifyZKP-AIRPQC ArchitecturePQ VideoStorage EncryptionAI DetectionEncrypted Search
Encrypted computation

Fully Homomorphic Encryption (FHE) API

Compute on encrypted data without ever decrypting it. H33's FHE API ships BFV and CKKS engines with FHE-IQ auto-routing — purpose-built in Rust, not wrapped around an open-source library.

943 μs
32-User Batch
42 μs
Per Auth
0 bytes
Plaintext Window
74 B
Attestation Size
What is FHE

Compute on Encrypted Data. Get Encrypted Results.

Fully Homomorphic Encryption lets you perform arbitrary computation on ciphertext. The server processes your data without ever seeing it. The result is returned encrypted — only you can decrypt it.

Think of it like a glove box: you manipulate the contents through sealed gloves. The materials are processed. The box never opens. With FHE, the "box" is mathematics, and opening it requires solving problems no computer (classical or quantum) can solve efficiently.

Traditional encryption protects data at rest and in transit. FHE protects data during computation — the last gap in the data lifecycle.

The difference

Purpose-Built, Not Wrapped

Most FHE offerings are thin API layers over Microsoft SEAL or OpenFHE. H33's FHE engines are written from scratch in Rust with production-specific optimizations:

Montgomery NTT — no division in the hot path
Harvey lazy reduction — deferred modular reduction between butterfly stages
SIMD batching — 32 users per ciphertext via CRT slot packing
NTT-domain persistence — skip redundant forward/inverse transforms
Batch attestation — 1 Dilithium sign per 32-user batch (not per user)

The result: 943 μs per 32-user batch on production hardware (Graviton4 metal-48xl, 192 vCPUs).

Engines
Three Engines. One Unified API.
Choose your engine explicitly, or let FHE-IQ auto-select based on your workload.

H33-128 (BFV)

Exact integer arithmetic

Bit-perfect computation on encrypted integers. Ideal for biometric matching, encrypted search, and any workload requiring exact results.

Parameters: N=4096, single 56-bit modulus, t=65537. NIST Level 1 security.

Batch
943 μs
Users/CT
32
Security
RLWE
PQ-Safe
Yes

H33-CKKS

Approximate floating-point

Approximate arithmetic on encrypted floating-point values. Purpose-built for ML inference, statistical analysis, and neural network evaluation on encrypted data.

Supports Chebyshev polynomial activations for encrypted neural network layers.

Type
Float
Precision
~40-bit
Security
RLWE
PQ-Safe
Yes

FHE-IQ

Intelligent auto-routing

Analyzes your workload and automatically selects the optimal FHE engine. Examines data type, precision requirements, multiplicative depth, and target latency.

Call one endpoint — FHE-IQ handles engine selection, parameter tuning, and batching strategy.

Mode
Auto
Engines
BFV+CKKS
Routing
Adaptive
PQ-Safe
Yes
Key differentiator
Attest FHE Results Without Decrypting

The Problem With Every Other FHE System

FHE computes on encrypted data. But to sign the result — to prove it was computed correctly — traditional signature systems require plaintext. This forces decryption, creating a plaintext window. The data you encrypted to protect is now exposed.

H33-74 solves this. It commits to the encrypted result directly, hashes it under SHA3-256, and signs the hash under three PQ signature families. The result: a 74-byte attestation that proves computation integrity without ever decrypting the data.

Learn more: How to Sign Encrypted Computation Without Decrypting It

Benchmarks
Production Performance
All numbers from Graviton4 metal-48xl (192 vCPUs, 371 GiB). Full pipeline including FHE, attestation, and ZKP verification.
Metric Value Notes
Sustained throughput 2,216,488 auth/sec 30s sustained, full pipeline
FHE batch (32 users) 943 μs BFV inner product, SIMD slots
Per-auth latency 42 μs Including attest + ZKP
Batch attestation 391 μs SHA3 + Dilithium sign + verify
ZKP cached lookup 0.358 μs CacheeEngine in-process
Per-auth cost ~$3.8 × 10-10 At on-demand pricing
API example
Encrypt and Compute in Two Calls
// 1. Encrypt data client-side
POST /v1/fhe/encrypt
{
  "engine": "bfv",
  "data": [128, 64, 32, ...]
}

// 2. Compute on encrypted data
POST /v1/fhe/compute
{
  "ciphertext": "<base64-ct>",
  "operation": "inner_product",
  "attest": true
}

// Response: encrypted result + H33-74 attestation
{
  "result_ct": "<base64-encrypted-result>",
  "h33_74": "<74-byte-substrate>",
  "compute_ms": 0.943
}
Explore
Related Pages