The complete technical story behind H33 — the cryptographic engines, NIST security levels, the full post-quantum lifecycle, performance benchmarks, API services, and verification standards.
Every measured number on this page is clickable to the proof behind it. For the product overview, pricing, and getting started, see the main homepage.
Post-Quantum, End-to-End — From NIST Level 1 to Level 5
Don't build post-quantum primitives yourself. H33 exposes the full PQ, FHE, and ZK stack through one authenticated interface. Same primitives. Same audit posture. Production-ready.
One Unified Interface
Identity & access — sign in, H33-Key
Products & services — discovery
Proof & verification — replay evidence
AI governance — governed agents
Routed by your authenticated identity, entitlements, and available capabilities. Nothing to install or choose between.
Behind an API
curl -X POST https://api.h33.ai/v1/sign \
-H "X-API-Key: $KEY" \
-d '{"alg":"ml-dsa-87","payload":"..."}'
No SDK required. Same primitives, exposed via HTTPS. Authenticated. Versioned. Specified.
The same cryptographic stack behind H33's portable proofs — now exposed for your custom work.
The Architecture · One System
The Full Cryptographic Lifecycle.
Eight Layers · One Continuously Attestable Pipeline
H33 continuously preserves institutional cryptographic state — from
generation at origin, through
encrypted compute,
governance,
attestation,
anchoring, and
independent replay.
Enables verifiable tokenization of real-world assets with preserved provenance, authority lineage, privacy, and chain portability.
13 Pending Patent Applications · 250+ Claims · One Architecture
Encrypted AI · Verifiable Infrastructure
Built for encrypted AI, post-quantum security, and verifiable enterprise infrastructure.
H33 systems produce cryptographically verifiable audit trails for AI systems, enterprise infrastructure, financial systems, and sensitive data workflows.
Encrypted computation produces results. Signature systems need plaintext. Decrypting to sign exposes data.
Unverifiable Pipelines
Multi-step AI and data pipelines have no cryptographic chain of custody. Every handoff is unsigned.
Lost Provenance
Data flows through systems with no proof of who attested it, how it was routed, or what computation ran.
Traditional logs cannot independently prove what AI systems and enterprise infrastructure actually did.
Most organizations expose sensitive enterprise and AI data during computation, then rely on logs, screenshots, and SIEM alerts after incidents occur. H33 prevents plaintext exposure during AI and enterprise computation while continuously generating replayable cryptographic evidence.
H33 closes all four gaps in a single pipeline.
What happens when AI has no cryptographic accountability
TRADING
A trading agent executes a $50M order and claims it received different instructions.
MEDICAL
A medical AI recommends a treatment but can't prove what data it was given.
LEGAL
A legal AI drafts a contract — the output is modified before the lawyer sees it.
FRAUD
A fraud detection system is fed manipulated inputs and produces a clean result.
AUTONOMOUS
An autonomous system causes harm and points to the previous agent in the chain.
Most AI systems cannot independently prove how decisions were made, what data was processed, or whether operational controls actually executed. As AI gains authority inside enterprises, the ability to independently reconstruct operational behavior is disappearing. H33 provides encrypted AI processing, AI audit trails, compliance verification, fraud prevention infrastructure, and cryptographic operational replay.
The Trailer · Five Subsystems · One Machine
Inside the Machine.
Five interoperating subsystems. Multiple
patented systems. One
replayable operational infrastructure layer.
From origin capture to independent verification.
Stage 01
INPUTS
IoT & Sensors
AI Models
Humans
Databases
Agents
Devices
Five subsystems · Nine patent applications on this architecture alone · One continuously attestable pipeline
How It Works
From Encrypted AI Processing to Cryptographic Verification
Every AI and enterprise decision becomes cryptographically verifiable audit evidence.
H33 continuously generates cryptographically verifiable audit trails for AI systems, enterprise infrastructure, financial systems, and sensitive operational environments — capturing what executed, what governance rules were active, what runtime produced the result, and whether the outcome can be independently verified later.
1
Data enters
Raw input encrypted on-device before transmission
2
Constraints checked
Domain separation registry validates computation type
3
Routed to engine
FHE-IQ selects optimal engine: BFV, CKKS, TFHE, or BFV-256
4
Computation executes
Encrypted inner product, rotation, threshold decrypt📐🔑 Interactive Shamir demo → — all in FHE
Bitcoin, Solana, Ethereum, database, or any verifiable store. Chain-agnostic.
Total pipeline latency: 1.36ms📐📐 Full benchmark suite → (H33-128📐🔒 128-bit NIST L1 →) · Zero plaintext exposure at any stage.
H33-74
A Small Attestation. A Much Bigger Economic Advantage.
The portable proof primitive powering H33 AIR.
Every H33 engine exports trust through the same primitive: H33-74 — a fixed 74-byte portable attestation, regardless of the size of the evidence it references.
That means organizations can carry a compact cryptographic reference across systems while retaining the underlying evidence separately. At enterprise scale, reducing the information repeatedly transmitted, anchored, or replicated can have significant operational and economic value.
The breakthrough isn’t simply that the attestation is small. It’s what that compact structure makes possible.
Operational and economic value depends on a given workload. H33 publishes no total-cost or storage measurement, and the 74 bytes is the attestation — not the signatures or the evidence package, which are carried separately. See what is and is not counted →
Fixed
74 bytes regardless of how large the evidence it references is
Portable
Moves through APIs, storage, devices, ledgers
Post-Quantum
Built for family rotation and long-lived verification
Independent
Verification does not depend on trusting H33
Continuity
Evidence survives cloud migration, software replacement, chain migration, model replacement, and vendor change
Sensitive enterprise and AI data remains encrypted throughout processing and verification. The computation happens inside the encryption. The result is post-quantum signed at the moment it's produced. The attestation stays 74 bytes however large the evidence it commits to becomes — so what a system must move and anchor does not grow with it.
Small proof. Massive accountability.
Same 74 bytes. Six simultaneous roles. All networked.
Most digital asset systems assume today's blockchain will still be the blockchain of choice ten years from now.
H33-74 separates evidence from any single chain. Assets, approvals, attestations, and audit evidence can be independently verified and re-anchored without losing provenance.
Transaction: 7f8d9ef2...b028b4a7 — Taproot P2TR + OP_RETURN. Standard Bitcoin. No soft fork. No new opcodes. Independently verifiable by any full node worldwide.
Post-Quantum Cryptography and Encrypted Computation API📐⚙️ Proprietary engine deep-dive →
One API integrates encrypted AI processing, cryptographic verification, post-quantum signing, compliance verification, and replayable audit infrastructure into a single production pipeline.
H33 combines post-quantum security, encrypted AI infrastructure, cryptographic audit trails, and continuous verification into a single API platform — enabling encrypted AI inference, post-quantum authentication, secure transaction verification, and zero-trust operational integrity.
Authentication Decisions
FHE · ZK · Biometrics
Computation Decisions
BFV · CKKS · TFHE · FHE-IQ
Verification Decisions
STARK Lookup📐🛡️ STARK lookup breakdown → · ZKP-AIR
Multi-Party Decisions
MPC · Agent-74 · PQ Video
All four export trust through H33-74.
AI Blind processes your data without ever seeing it. H33-74 attests the result without decrypting it.
Performance
Fast Enough for Inline AI Security and Enterprise Infrastructure📐📐 Full benchmark suite →
Not a batch job. Not an async queue. Inline, synchronous, every request.
Mode
Latency
Security
Exposure
H0 (Classical)
0.8ms
None (quantum-breakable)
Full plaintext
H33-128
1.36ms📐📐 Full benchmark suite →
NIST L1📐🔒 NIST security levels →
Zero
H33-256
5.98ms📐📐 Full benchmark suite →
NIST L5📐🔒 NIST security levels →
Zero
Why H33 Is Faster Than Expected
⚙️
Purpose-Built Engines
Every FHE and ZK engine written from scratch. No wrappers around SEAL, OpenFHE, or circom. The math is ours.
🔗
Zero External Dependencies
No third-party FHE or ZK libraries in the critical path. Every gate, every polynomial, every proof — compiled from our source.
🔄
Pipeline-Level Optimization
FHE encrypt, compute, attest, and sign happen in a single fused pipeline. No serialization between stages. No redundant transforms.
📦
Fixed-Size Trust Export
Every operation produces a fixed 74-byte attestation: a 32-byte commitment that anchors on-chain plus a 42-byte companion component. Signatures and the underlying evidence travel alongside it, not inside it.
A human eye blink takes ~300ms. In that time, H33-128 completes 220full pipeline authentications.
⚡H33 vs Microsoft SEAL5-25x faster▼
Operation
SEAL
H33
Factor
BFV Encrypt📐🔧 Single-modulus explainer →
~2.1ms
0.42ms
5x
Multiply
~6.5ms
0.26ms
25x
Full Pipeline📐📐 Full benchmark suite →
N/A
1.36ms
Unique
SEAL does not include ZK proofs📐🛡️ STARK lookup breakdown →, PQ signatures📐✍️ Batch attestation demo →, or biometric matching📐🧬 Biometrics engine deep-dive →. H33 includes all three in 1.36ms.
The Stack Is Ours.📐⚙️ Proprietary engine deep-dive →
Every engine, every prover, every signature library — built from scratch. No wrappers. No forks. No external FHE or ZK dependencies📐⚙️ Proprietary engine deep-dive →.
One Platform. Multiple AI Security and Verification Systems.
Every product below is built on the same 74-byte trust primitive.
H33 enables verifiable AI systems with encrypted processing, post-quantum signatures, deterministic replay, and independently verifiable operational evidence.
Eleven FHE library members with intelligent routing. BFV adds and multiplies. CKKS runs encrypted ML workloads. TFHE decides. All post-quantum attested.
Existing systems produce logs. H33 produces replayable evidence — every decision can be reconstructed, verified, and tied back to the exact runtime that produced it.
Verification Standards
One Cryptographic Lifecycle. Multiple Verification Layers.
We use cookies to enhance your browsing experience, analyze site traffic, and personalize content. By clicking "Accept All", you consent to our use of cookies. You can manage your preferences or learn more in our Cookie Policy.
Cookie Preferences
Required for the website to function. Cannot be disabled.
Help us understand how visitors use our website.
Used to deliver relevant advertisements.
Remember your settings and preferences.
Live Production Test
1,000 Real Auths. Your Browser. Right Now.
Every request hits our production pipeline: FHE biometric + ZK STARK (Cachee pre-cached) + Dilithium attestation on ARM Graviton. Full pipeline, real crypto.
H33-128us-east-1Detecting...
Run 1,000 live post-quantum auths
We’ll email you the full performance report so you can share it with your team.
Free · No API key required · Results + report delivered to your inbox
Solving proof-of-work challenge...
0K hashes
Cryptographic verification — takes a few seconds
0 / 1,000
Current Server
--
Avg Server
--
Network
--
Avg RTT
--
⏳
Demo Limit Reached
You've used all 3 demo runs for this hour. Every demo run executes real FHE, ZK, and Dilithium operations on our servers.
Single instance demo. You just ran the full production pipeline (FHE + ZKP + Dilithium) on one ARM core. Production deploys 96 parallel workers on Graviton4 metal — 2,293,766 auth/sec sustained, 35.25µs per auth.📐📐 See how we measured this → Same code, same optimizations, more cores.
Free Demo
Full post-quantum pipeline — FHE + ZKP + Dilithium, no shortcuts. SIMD batching packs 32 users into one ciphertext. 2,293,766 auth/sec sustained. 35.25µs per auth.
Both. They're two views of the same H33-74 artifact — you'll see whichever number is most relevant to what's being measured.
32 bytes
The commitment
A post-quantum commitment that anchors on-chain. This is the verification footprint — what an auditor independently checks, years later, with no H33 infrastructure required.
74 bytes
The attestation
The full portable artifact: 32-byte commitment + 42-byte envelope (lineage and version metadata), signed under three NIST post-quantum families. The signatures and the public keys needed to check them travel alongside the attestation, not inside it — a single post-quantum signature is itself far larger than 74 bytes. What flows through your systems.
Your blockchain stores the 32 bytes. Your auditors and verifiers consume the 74. Same artifact, different boundary.
