PricingDemo
Log InGet API Key
Boolean FHE Engine

Encrypted logic.
Gate by gate.

AND, OR, XOR, NOT — on data that is never decrypted. Every gate operates on ciphertext. The server processes logic without seeing values, identities, or inputs.

11,526
raw gates/sec · single-bit AND gates · 96 channels · Graviton4 ARM · no GPU
Get API Key TFHE Bootstrap

The smallest unit of
encrypted computation

Every encrypted decision breaks down to gates. AND, OR, XOR, NOT — these are the atoms. H33-TFHE evaluates these gates directly on LWE ciphertexts.

No polynomial arithmetic. No floating point. No approximation. Exact Boolean logic on encrypted bits.

This is what runs underneath every comparison, every threshold check, every encrypted decision.

Gate operations

Seven gates. The building blocks of every encrypted circuit.

Logic Gates

  • AND Fundamental gate, 1 bootstrap
  • OR Logical disjunction, 1 bootstrap
  • NAND Universal gate, 1 bootstrap
  • NOR Universal gate, 1 bootstrap
  • XOR No noise cost FREE
  • XNOR No noise cost FREE
  • NOT Bit flip, zero cost FREE

What you build with gates

  • Comparators — greater-than, less-than, equality
  • Adders and subtractors — encrypted arithmetic from logic
  • Multiplexers — encrypted routing
  • Threshold checks — is value above or below limit?
  • Decision trees — encrypted if/else chains

Gates are primitives. Circuits are products.

Measured Performance

All numbers measured on AWS Graviton4 (ARM). No GPU. Sustained throughput.

GateCostThroughput (96ch)
AND1 bootstrap11,526/sec
OR1 bootstrap11,526/sec
NAND1 bootstrap11,526/sec
NOR1 bootstrap11,526/sec
XORFreeNo limit
XNORFreeNo limit
NOTFreeNo limit

Measured on AWS Graviton4 (ARM). No GPU. Sustained throughput. Free gates (XOR, XNOR, NOT) require no bootstrapping — they operate on ciphertext noise-free.

Where TFHE fits in the stack

H33 runs five FHE engines. Each one solves a different problem. FHE-IQ routes between them automatically.

BFV
Batched integer arithmetic. Millions of ops/sec.
Best for: biometric matching, batch auth
BFV-256
Higher security margin. Same pattern, stronger parameters.
Best for: classified workloads
CKKS
Approximate arithmetic on encrypted floats.
Best for: ML inference, regression, neural nets
H33-TFHE
Boolean gate logic on encrypted bits.
Best for: binary decisions, comparisons, control flow
You are here
TFHE Bootstrap
Unlimited-depth circuits via programmable bootstrapping.
Best for: policy engines, complex encrypted decisions

FHE-IQ routes automatically between them. You send data. The system selects the engine.

Fixed depth vs
unlimited depth

H33-TFHE (this page)

Fixed circuit depth. You design the circuit, noise budget is set by parameters.

Fast for shallow circuits. Predictable cost per gate.

11,526 bootstrapped gates/sec. Free XOR/NOT operations.

Use TFHE when your circuit is bounded.

TFHE Bootstrap

Every gate resets noise via programmable bootstrapping. No depth limit.

More expensive per gate, but no circuit planning required.

768 TPS for 8-bit comparisons. Unlimited composition.

Use TFHE Bootstrap when your circuit is not.

Use TFHE when your circuit is bounded. Use TFHE Bootstrap when it is not.

See TFHE Bootstrap →

Post-quantum secure

TFHE is based on lattice cryptography — the Learning With Errors (LWE) hardness assumption. The same class of mathematical problem that underpins all NIST post-quantum standards.

Every gate result can be attested via H33-74 — 74 bytes, distilled from three independent hardness assumptions: MLWE lattices, NTRU lattices, and hash-based signatures.

Not just encrypted. Provable.

1. Input bits encrypted as LWE ciphertexts
2. Gates evaluated on ciphertext — server never sees values
3. Output bits remain encrypted
4. Result attested — three independent hardness assumptions
5. Proof committed — 74 bytes, permanent
Verify It Yourself