PQ Crypta Logo

What is PQ Crypta?

The First Post-Quantum Operational Intelligence Platform

๐ŸงญDefining a New Category: PQ-OIP

PQ Crypta defines a new category of security infrastructure: the Post-Quantum Operational Intelligence Platform (PQ-OIP) โ€” a unified environment that validates, observes, and operationalizes post-quantum cryptography across real systems, real protocols, and real deployments.

A PQ-OIP is not a library. Not a PKI vendor. Not a migration consultancy. It is a living operational substrate that gives engineering teams the same capabilities observability platforms gave SREs: continuous insight, reproducible validation, and actionable intelligence for a cryptographic world undergoing irreversible change.

What Makes This a New Category?

Existing players occupy narrow buckets:

  • PQC Libraries (liboqs, pqm4) โ†’ Code, not operations
  • PKI Vendors โ†’ Certificates, not protocol behavior
  • Crypto-Agility Platforms โ†’ Inventory, not validation
  • Consultancies โ†’ Documents, not reproducible tooling
  • Research Labs โ†’ Papers, not operational systems

None combine protocol scanning, operational observability, reproducible validation, and developer-centric tooling into a single platform. PQ Crypta is the first.

The Four Pillars of PQ-OIP

๐Ÿ”ฌ
Protocol-Level Validation
Side-channel-aware binary formats, hybrid handshake inspection, QUIC/TLS scanning, and deterministic test vectors
๐Ÿ“Š
Operational Observability
Crypto-health dashboards, algorithm lifecycle monitoring, entropy diagnostics, and failure-mode analysis
๐Ÿ› ๏ธ
Developer Migration Tooling
Rust-first implementations, cross-language specs, integration harnesses, and reproducible workflows
๐Ÿ“š
Living Knowledge Layer
RAG-augmented documentation, ecosystem mapping, adoption playbooks, and accessible education

๐Ÿ“‹Platform Overview

PQ Crypta delivers post-quantum operational intelligence through a dual-architecture system: interactive PHP/JavaScript applications for hands-on exploration, and a high-performance Rust API backend for production integration.

Core Mission: Provide continuous cryptographic observability and actionable intelligence for teams preparing for the post-quantum transition โ€” from protocol validation to production deployment.

691 REST API Endpoints
35 Validated Algorithms
96 Interactive Tools
99.9% Endpoints Healthy, Last Run

๐Ÿ—๏ธSystem Architecture Overview

Dual Architecture Design

1. Frontend Layer (PHP 8.4 + Vanilla JavaScript, ES2024)

  • 96 pages and interactive tools in the site menu
  • Zero-knowledge client-side cryptography
  • Real-time visualization (Three.js, WebGL, WebGPU)
  • Modern build system (Vite 8.3.7)
  • Served by Apache and PHP-FPM on loopback, behind the proxy (which speaks HTTP/1.1 to it, and HTTP/2 and HTTP/3 to browsers)

2. Backend API Layer (Rust + Axum Framework)

  • 691 REST API endpoints (446 static + 245 algorithm-specific)
  • Engines on the final standards (FIPS 203, 204 and 205, and FN-DSA): round-3 Dilithium, SPHINCS+ and Falcon signatures are refused, not accepted for compatibility
  • High-performance async runtime (Tokio)
  • Enterprise business logic services
  • PostgreSQL database integration
  • Deployed on Rust API server (port 3003)

Endpoint Breakdown:

  • 446 static endpoints (core services, analytics, auth, attestation ledger, discovery, scanners, etc.)
  • 245 algorithm-specific endpoints across 35 algorithms (7 core operations each)
  • Operations per algorithm: POST /keys/generate/:algorithm, /encrypt/:algorithm, /decrypt/:algorithm, /validate/:algorithm, /info/:algorithm, /benchmark/:algorithm, /algorithms/info/:algorithm

3. Reverse Proxy Layer (pqcrypta-proxy โ€” custom Rust)

  • Purpose-built Rust reverse proxy (rustls on QUIC through noq, OpenSSL 4.0.3 on TCP) deployed on 5 nodes
  • HTTP/3 with QUIC, HTTP/2, WebTransport (UDP 443), and 103 Early Hints over HTTP/3 and HTTP/2
  • TLS 1.3 with post-quantum hybrid key exchange (X25519MLKEM768) and integrated ACME certificate management
  • Live pure-PQ TLS endpoint: pqc.pqcrypta.com serves an ML-DSA-87 certificate chain rooted in SLH-DSA
  • WAF, JA3/JA4 TLS fingerprinting, GeoIP country blocking, rate limiting, bot protection, honeypot pipeline
  • Response caching, load balancing, circuit breakers, and traffic routing between frontend (pqcrypta.com) and API (api.pqcrypta.com)

4. Database Layer (PostgreSQL 16.15)

  • Comprehensive analytics tracking
  • Health monitoring with 30-day retention
  • GDPR-compliant dual-mode tracking (anonymous + consented)
  • Cryptographic asset discovery inventory
  • The attestation ledger's index: final blocks and transactions, searchable by digest, key, signer and kind, rebuilt from the chain the validators keep
  • User authentication and session management

5. ML/AI Layer (Python 3.12.3; openai/gpt-oss-120b on Groq for the AI regex generator)

  • The Wizard AI Chatbot with RAG over a FAISS vector store โ€” disabled since 2026-07-06 and not compiled into the API
  • Compression's auto mode chooses by measurement, not by prediction: a sample of the input compressed with each candidate codec, the smallest kept
  • HTTP/3 recommendation model (scikit-learn random forests, 100 trees each), trained on the analyzer's own scans
  • Federated threat model: trained with differential privacy on api3, mail, ns4 and web1's own traffic, which never leaves the node
  • Bot classifier (gradient-boosted trees) trained on the site's own traffic, finding the automated visitors the rules miss

6. Automation Layer (Cron Jobs + Scripts)

  • 58 scheduled jobs on systemd timers (listed below), among them the health check, every 15 minutes (691 endpoints, real cryptographic workflows)

๐Ÿ”35 Validated Algorithms

All algorithms verified operational with 100% success rate across full cryptographic workflows (Key Generation โ†’ Encryption โ†’ Decryption โ†’ Validation). Every engine performs real KEM encapsulation and signature verification, checked every 15 minutes by the health check with tamper-rejection tests (modified ciphertexts are refused with HTTP 422).

Each engine below is key establishment | signatures | encryption, as it declares them; its FIPS standing is on the compliance page.

๐Ÿงฑ Core (5)

X25519 alone (classical); ML-KEM-1024 with ML-DSA-87 (post-quantum); both, combined by HKDF-SHA256 over the secrets, ciphertexts and keys (hybrid); ML-KEM-1024 alone; and multi-algorithm: ML-KEM-1024 and X25519 under a KMAC256 combiner, signed with ML-DSA-87 and SLH-DSA-SHA2-256s, with AES-256-GCM inside XAES-256-GCM.

  • Classical (classical): X25519 | Ed25519 | AES-256-GCM
  • Hybrid (hybrid): ML-KEM-1024, X25519 | ML-DSA-87, Ed25519 | AES-256-GCM
  • Post-Quantum (post-quantum): ML-KEM-1024 | ML-DSA-87 | AES-256-GCM
  • ML-KEM-1024 (ml-kem-1024): ML-KEM-1024 | AES-256-GCM
  • Multi-Algorithm (multi-algorithm): ML-KEM-1024, X25519 | ML-DSA-87, SLH-DSA-SHA2-256s | AES-256-GCM, XAES-256-GCM

๐Ÿ”— Multi-KEM and Layered (5)

Several KEMs whose secrets one SP 800-227 key combiner (the KMAC256 one-step KDF) turns into the key: reading the data means breaking every KEM in it. PQ3 Stack also runs as a post-quantum double ratchet with forward secrecy (/pq3/session/*).

  • Multi-KEM (multi-kem): ML-KEM-1024, X25519 | SLH-DSA-SHA2-256s | AES-256-GCM
  • Multi-KEM Triple (multi-kem-triple): ML-KEM-1024, ML-KEM-1024, X25519 | SLH-DSA-SHA2-256s | AES-256-GCM
  • Quad-Layer (quad-layer): ML-KEM-1024, X25519 | ML-DSA-87, Ed25519, SLH-DSA-SHA2-256s | AES-256-GCM
  • PQ3 Stack (pq3-stack): ML-KEM-1024, X25519 | ML-DSA-87 | AES-256-GCM
  • Lattice-Code Hybrid (lattice-code-hybrid): ML-KEM-1024, Classic-McEliece-348864 | SLH-DSA-SHA2-256s | AES-256-GCM

๐Ÿ›ก๏ธ Max Secure (6)

Category 5 stacks for particular needs. Lightweight runs in the browser too; crypto-agile records AES-256-GCM or XAES-256-GCM per key pair; PQC + ZK proves a plaintext commitment with a zk-STARK.

  • Max Secure Lightweight (max-secure-lightweight): ML-KEM-1024, X25519 | Ed25519 | AES-256-GCM
  • Max Secure Pure PQ (max-secure-pure-pq): ML-KEM-1024 | AES-256-GCM
  • Max Secure Hybrid Transition (max-secure-hybrid-transition): ML-KEM-1024, X25519 | ML-DSA-87, Ed25519 | AES-256-GCM
  • Max Secure Stateless (max-secure-stateless): ML-KEM-1024 | SLH-DSA-SHA2-256s | AES-256-GCM
  • Max Secure Crypto-Agile (max-secure-crypto-agile): X25519, ML-KEM-1024 | Ed25519, ML-DSA-87, FN-DSA-1024, SLH-DSA-SHA2-256s | AES-256-GCM, XAES-256-GCM
  • Max Secure PQC + ZK (max-secure-pqc-zk): ML-KEM-1024 | SLH-DSA-SHA2-256s, zk-STARK | AES-256-GCM

๐Ÿ” FN-DSA (6)

FN-DSA as the FIPS 206 draft specifies it (fn-dsa), every signature a fixed size: 666 bytes for FN-DSA-512, 1280 for FN-DSA-1024. FP-Hardened signs on integer-emulated floating point (no FPU).

  • FN-DSA 512 Compact (fn-dsa-512-compact): ML-KEM-1024 | FN-DSA-512 | AES-256-GCM
  • FN-DSA 1024 Security (fn-dsa-1024-security): ML-KEM-1024 | FN-DSA-1024 | AES-256-GCM
  • FN-DSA FP-Hardened (fn-dsa-fp-hardened): ML-KEM-1024 | FN-DSA-1024 | AES-256-GCM
  • FN-DSA Dual-Signature (fn-dsa-dual-signature): ML-KEM-1024 | FN-DSA-1024, SLH-DSA-SHA2-256s | AES-256-GCM
  • FN-DSA Transition Stack (fn-dsa-transition-stack): ML-KEM-1024, X25519 | FN-DSA-512, Ed25519 | AES-256-GCM
  • FN-DSA ZK Stack (fn-dsa-zk-stack): ML-KEM-1024 | FN-DSA-512, Groth16 | AES-256-GCM

๐Ÿ“ก HQC Code-Based (3)

HQC, which NIST selected in March 2025 as a second, code-based KEM beside ML-KEM; it awaits its final standard.

  • HQC-128 (Level 1) (hqc-128): HQC-128 | Ed25519 | AES-256-GCM
  • HQC-192 (Level 3) (hqc-192): HQC-192 | Ed25519 | AES-256-GCM
  • HQC-256 (Level 5) (hqc-256): HQC-256 | Ed25519 | AES-256-GCM

๐ŸŒ FIPS-Final / ISO KEMs (4)

X25519MLKEM768 pairs the two key exchanges of TLS’s hybrid, ML-KEM-768 and X25519, under X-Wing’s SHA3-256 combiner, which SP 800-227 approves. Classic McEliece (ISO/IEC 18033-2) and FrodoKEM (unstructured LWE, ISO/IEC 18033-2 track, cited in BSI TR-02102-1) are conservative KEMs NIST has not approved, so their engines report them as not approved.

  • X25519MLKEM768 (x25519-mlkem768): ML-KEM-768, X25519, X-Wing | AES-256-GCM
  • Classic McEliece 6960119 (classic-mceliece): Classic-McEliece-6960119 | AES-256-GCM
  • FrodoKEM-976 (Level 3) (frodokem-976): FrodoKEM-976 | AES-256-GCM
  • FrodoKEM-1344 (Level 5) (frodokem-1344): FrodoKEM-1344 | AES-256-GCM

โš—๏ธ Experimental (5)

Post-ZK-Homomorphic encrypts under TFHE (tfhe-rs integer, KS32-PBS) as well, so /fhe/evaluate computes on ciphertexts with the server key alone; Quantum-Lattice-Fusion pairs ML-KEM-1024 with FrodoKEM-1344 under HKDF-SHA256.

  • Quantum-Lattice-Fusion (quantum-lattice-fusion): ML-KEM-1024, FrodoKEM-1344 | FN-DSA-1024 | AES-256-GCM
  • Post-ZK-Homomorphic (post-zk-homomorphic): ML-KEM-1024 | FN-DSA-1024, zk-STARK | AES-256-GCM, TFHE
  • Quantum-Resistant-Consensus (quantum-resistant-consensus): ML-KEM-1024 | FN-DSA-1024 | AES-256-GCM
  • Entropy-Orchestrated (entropy-orchestrated): ML-KEM-1024 | FN-DSA-1024 | AES-256-GCM
  • AI-Synthesized-Crypto-Agile (ai-synthesized-crypto-agile): ML-KEM-1024, X25519 | ML-DSA-87, SLH-DSA-SHA2-256s | AES-256-GCM

๐Ÿ”‘ Password-Based (1)

Keys from a passphrase with Argon2id (RFC 9106, 64 MiB, t = 3, p = 4): memory-hard, where PBKDF2, the FIPS-approved choice, is not, so the engine reports Argon2id as not approved.

  • Password Classical (password-classical): X25519 | Ed25519 | AES-256-GCM

Post-ZK-Homomorphic in detail

  • TFHE (tfhe-rs integer, KS32-PBS, 2-128) under ML-KEM-1024 + AES-256-GCM, signed with FN-DSA-1024, with a zk-STARK proof of the plaintext commitment
  • Computes on encrypted integers: /fhe/evaluate adds, subtracts, multiplies, compares and selects with the server key alone, never decrypting
  • Keys: 106 MB seeded server key (281 MB expanded), kept in the key vault; a u32 addition takes 390 ms, a product 3.9 s, measured on the API host, 4 cores (examples/fhe_params.rs)
  • Use case: computation delegated to a server that never sees the data

Measured speed (key generation + encryption + decryption through the API, median of five full benchmark runs; the full distribution):

  • Fastest: X25519MLKEM768 36 ms, ML-KEM-1024 45 ms, Classical 47 ms
  • Slowest: Quad-Layer 875 ms, Lattice-Code-Hybrid 999 ms, Post-ZK-Homomorphic 3540 ms
  • Key Size Range: Classical (~64 bytes) โ†’ Post-ZK Homomorphic (106 MB seeded TFHE server key)

๐ŸŒFrontend Applications (96 Pages)

PQ Crypta features an extensive collection of interactive web applications spanning encryption, compression, security scanning, AI tools, visualizations, and entertainment.

1. Encryption Suite (/encryption/index.php)

Purpose: Zero-knowledge client-side encryption with 35 algorithms (33 quantum-resistant)

  • Dual Modes: Password Mode (simple) & Key Mode (advanced)
  • 35 Total Algorithms: 33 quantum-resistant, plus Classical and Password-Classical
  • File Support: any file, of any size or kind, encrypted whole in the page: read and written a chunk at a time (AES-256-GCM in the STREAM construction, to the browser's own disk storage), its key encrypted by the selected algorithm or the password; a text message up to 10,000 characters as before
  • Security: All operations client-side; keys never leave device
  • Technology: @noble/post-quantum, @noble/curves, Vite.js, WebAssembly

2. Compression Platform (/compression/index.php)

Purpose: Data compression with 10 codecs, and an auto mode that measures rather than guesses

  • 10 Codecs: zstd, lz4, brotli, deflate, gzip, snappy, xz, bzip2, ppmd, context-mixing
  • Auto: a sample of the input compressed with each of zstd, brotli, xz, bzip2, ppmd, and the smallest kept
  • PQCZ Format: PQCrypta Compressed format with metadata headers

3. PQC Readiness Scanner (/pqc-ready/index.php)

Purpose: A website's post-quantum readiness, measured over TLS and QUIC

  • Grading System: A+ (a browser gets standard ML-KEM, TLS 1.3 only, post-quantum key exchange even when a client lists a classical group first, the Kyber drafts refused), A (a browser gets standard ML-KEM, TLS 1.3 only), B (a browser gets standard ML-KEM, older TLS open), C (the server accepts ML-KEM but a browser does not get it), F (no post-quantum key exchange a browser can use), each grade's requirements shown met or not
  • Every Group: each TLS 1.3 key-exchange group offered on its own (standard ML-KEM, the obsolete Kyber drafts, experimental KEMs, classical), the server's order and the TLS 1.3 suites
  • Certificates: X.509 parsing, and a handshake asking for an ML-DSA signature to find an ML-DSA certificate
  • Beyond TCP: the key exchange over HTTP/3, Encrypted Client Hello acceptance, NSA CNSA 2.0 requirement by requirement
  • Directory: every scan recorded, a domain's record over time, and the figures across every domain measured; the Post-Quantum Web Analyzer makes the same assessment and records it there

4. Post-Quantum Web Analyzer (/http3-quic/index.php)

Purpose: Every layer a browser goes through to reach a site, DNS, TLS, QUIC and HTTP/3, measured and pushed, and its post-quantum readiness

  • One Grade: A++ to F for the whole scan, from adoption (the weighted share of what the scan looks for that the site has deployed, weighted core 6, hardening 4, frontier 2, capability 1) and conformance (the share of the RFC checks on what it deployed that it passes, area by area); A++: at least 95% adopted and 98% conformant, with post-quantum key exchange over TLS on TCP and over HTTP/3; A+: at least 85% adopted and 95% conformant, with post-quantum key exchange over TLS on TCP and over HTTP/3; A: at least 70% adopted and 90% conformant, with post-quantum key exchange over TLS on TCP and over HTTP/3; B: at least 55% adopted and 80% conformant; C: at least 40% adopted and 70% conformant; D: at least 25% adopted and 50% conformant; F: short of D: less than 25% adopted or 50% conformant. A++ needs at least 95% adopted. Where every item applies the weights total 67, and one item missing leaves at most 91.0% for a core item and 94.0% for a hardening item: A++ needs every core and hardening item, no TLS version older than 1.3 among them, so a site that still accepts TLS 1.2 reaches A+ at best. Of the frontier and the capabilities it may miss items weighing 3 in all (a frontier item weighs 2 and a capability item weighs 1). Each grade's requirements are shown met, not met or undecided, and HTTP/3 served but not advertised, or advertised but unreachable, named as such
  • Certificate Analysis: subject/issuer/validity/key size/signature algorithm, explicit PQC and composite classical+PQC signature detection, chain length, root CA, OCSP stapling, must-staple, Certificate Transparency SCT count
  • RFC 9218 Extensible Priorities: active PRIORITY_UPDATE frame test, not just passive header inspection
  • Multipath QUIC probe: experimental capability check for the draft-ietf-quic-multipath-21 extension (informational, not graded โ€” still an active IETF draft)
  • Encrypted DNS at the name servers: each of the domain's name servers asked over DNS over QUIC (RFC 9250) and DNS over TLS (RFC 7858) on port 853 and DNS over HTTPS (RFC 8484) over HTTP/2 and HTTP/3, with the key exchange (post-quantum or classical), certificate trust and timing; up to 74 conformance checks per server, all 47 RFC 9250 and 28 RFC 8484 requirements mapped; RFC 9461 discovery records, reverse names, networks and SOA timers (not graded: its own card and recommendations)
  • What Each Browser Gets: Chrome's, Edge's, Firefox's and Apple's own TLS 1.3 offers replayed over TCP and QUIC with real key shares, the group the server picks from each, and how each browser finds the site's HTTP/3
  • Mail Exchangers: each MX host over SMTP's STARTTLS, with the key exchange it negotiates
  • Post-Quantum Readiness: every key-exchange group on every transport, the older TLS versions, an ML-DSA certificate, ECH and CNSA 2.0, recorded in the PQC directory: the PQC tab, and what the grade's post-quantum items read
  • ML Recommendations: random forests trained on the analyzer's own scans suggest protocol upgrades, beside the PQC Readiness Scanner's (94.9% accurate on 144 domains held out of training)

5. Hardware Entropy Generator (/entropy/index.php)

Purpose: Interactive cryptographic randomness generation with quality testing

  • 4 Entropy Sources: Hardware (CPU RDRAND), System (/dev/urandom), Quantum (ID Quantique Quantis), Mixed (XOR combination)
  • NIST Statistical Tests: Frequency, Block Frequency, Runs, Longest Run of Ones
  • Quality Scoring: 0-10 scale with comprehensive metrics

6. The Wizard AI Chatbot (/ent/ml/src/chatbot/qwen_chatbot.py)

Status: disabled since 2026-07-06. Its service, handlers and routes are behind the API's wizard build feature, off by default, so /wizard/* and /chatbot/* answer 404 and its models are not loaded.

Purpose: Codebase-aware AI assistant with RAG, comprehensive PQCrypta algorithm knowledge, and security filtering

  • AI Model: openai/gpt-oss-120b on Groq, then gemini-2.5-flash on Gemini
  • RAG System: FAISS vector store over the codebase's documentation
  • Security Features: Prompt injection detection, output filtering, risk scoring
  • Comprehensive Algorithm Knowledge: Embedded knowledge of all 35 PQCrypta algorithms

7. Additional Applications

  • Discovery Agent - Cross-platform cryptographic asset discovery (Linux/Windows/macOS): TLS/keystore/appliance scanning (F5 BIG-IP, Citrix NetScaler), JKS/PKCS12 inventory, and CBOM export (CycloneDX 1.6)
  • Encrypt & Share - Nested dual-KEM sharing links: HQC-256 wrapped around ML-KEM-1024, so both algorithms must fall before data is exposed
  • QUIC Speed Test - WebTransport-native speed and latency test served from 3 nodes
  • Live PQ-TLS Demo - pqc.pqcrypta.com: TLS terminated with an ML-DSA-87 certificate chain rooted in SLH-DSA
  • AI Regex Generator - Natural-language regex creation with instant testing, and code in 22 languages
  • PQC Benchmark & Analysis - Browser-side algorithm benchmarking with a published statistical analysis of real run data
  • Interactive Mouse Effects - Custom cursor animations and visual effects
  • Streaming Music Player - High-performance MP3 player with quantum-resistant encryption
  • WebTransport Streaming Test - HTTP/3 WebTransport testing with PQC encryption
  • PQ Crypta Ledger - The PQ Crypta attestation ledger: ML-DSA-87 signed transactions and blocks, Tendermint BFT consensus among proof-of-authority validators; transactions signed in the browser
  • Web Analytics Platform - Enterprise-grade analytics with GDPR compliance
  • PQC Challenge Mode - Educational cryptography puzzle platform
  • Threat Detection Dashboard - Real-time bot attack monitoring
  • Bot Threat Remediation - Automated bot mitigation tools
  • Contact Form - Secure contact form with anti-spam
  • Quantum Morphic Field Explorer - Interactive quantum field visualization
  • Educational Animations - Cosmic Fluid, Neural Network, Fractal Tree, Particle System, etc.
  • News Widget - Quantum computing and cryptography news aggregator
  • ...and more!

๐Ÿ”ŒAPI Architecture (691 Endpoints)

Endpoint Categories

Total Endpoints: 691

  • Static Endpoints: 446 (core services)
  • Algorithm-Specific Endpoints: 245 (35 algorithms, 7 core operations each)

Static Endpoint Breakdown (446 endpoints, by OpenAPI tag)

Category Count
Analytics 58
Encryption 49
Authentication 43
Utilities 41
Bot Threat 25
User Management 22
ML/AI 18
Crypto Discovery 16
Billing 15
Music Service 15
Compliance 13
Support 13
File Operations 10
Scanners 9
Key Vault 9
Ledger 9
Zero-Knowledge 9
Batch Processing 7
Discovery Agent 7
Core 6
Streaming 6
Verifiable Credentials 6
Decentralized Identity 5
Policy Engine 5
Signatures 4
Homomorphic Evaluation 4
Regex Tools 4
SIEM Export 4
Zero-Knowledge Access 4
Metrics 3
General Crypto 2
News Service 2
Threat Intel 2
Key Management 1

Algorithm-Specific Endpoints (245 endpoints)

7 core operations per algorithm ร— 35 algorithms = 245 endpoints

For each of the 35 algorithms (33 quantum-resistant):

  1. POST /keys/generate/:algorithm - Generate key pair
  2. POST /encrypt/:algorithm - Encrypt data
  3. POST /decrypt/:algorithm - Decrypt data
  4. POST /validate/:algorithm - Validate keys
  5. GET /info/:algorithm - Algorithm information
  6. POST /benchmark/:algorithm - Performance benchmark
  7. GET /algorithms/info/:algorithm - Detailed algorithm specs

Example URLs:

  • POST https://api.pqcrypta.com/keys/generate/hybrid
  • POST https://api.pqcrypta.com/encrypt/post-zk-homomorphic
  • POST https://api.pqcrypta.com/decrypt/ml-kem-1024
  • GET https://api.pqcrypta.com/info/max-secure-pqc-zk

๐Ÿ—„๏ธDatabase Architecture

PostgreSQL 16.15 Schema

Database: pqcrypta
User: pqcrypta_user

Core Tables

  • Health Check Tables - every run of the 691-endpoint health check, every 15 minutes, and each endpoint's daily uptime
  • Analytics Tables - Web analytics sessions, pageviews, events, errors (GDPR-compliant)
  • Crypto Discovery Tables - Cryptographic asset inventory
  • Ledger Tables - the chain's index: meta, blocks, transactions
  • Threat Detection Tables - Bot attack logs, whitelists, patterns
  • ML/AI Tables - the federated threat model's rounds and updates, the HTTP/3 recommendation patterns, the Wizard's conversations and security logs
  • Scanner Tables - PQC scan results, HTTP/3 scanner results
  • User Management Tables - Users, API keys with granular permissions, sessions, 2FA configs
  • News & Content Tables - News articles, categories
  • Configuration Tables - Feature flags, rate limits, system config

Database Performance

  • Connection Pooling: the pages' through PgBouncer 1.26.1, pooled by transaction, a backend closed after 60 s idle; the API's its own (deadpool-postgres over tokio-postgres)
  • Query Optimization: B-tree indexes on frequently queried columns
  • Retention Policies: one rule per table (data_retention_rules), enforced daily at 03:40: 1 kept 1 day, 1 kept 1 year, 4 kept 30 days and 25 kept 90 days; 16 kept as long as needed (audit trails, security records, and analytics that follow their session's 90 days)
  • Backup Schedule: database dumps every 6 hours; a BorgBackup archive daily at 07:30; copied off the host every 15 minutes

๐Ÿค–ML/AI Integration

Python ML Services

Location: /var/www/html/public/ent/ml/
AI Model: openai/gpt-oss-120b on Groq, then gemini-2.5-flash on Gemini for the Wizard, openai/gpt-oss-120b on Groq for the AI regex generator

1. The Wizard AI Chatbot (disabled; not compiled into the API)

  • RAG System: FAISS vector store over the codebase's documentation
  • Comprehensive Algorithm Knowledge: All 35 PQCrypta algorithms embedded in system prompt
  • Security Features: Prompt injection detection, output filtering, risk scoring (0.0-1.0)
  • Caching: its last 100 retrievals, in memory

2. Compression: chosen by measurement

  • Method: the input's first 256 KiB compressed with each of zstd, brotli, xz, bzip2 and ppmd, and the smallest kept (/ml/compress and the codecs' auto)
  • Why not a model: a trial on the input's own bytes is exact for that input; a classifier would only predict what the trial measures

3. HTTP/3 Recommendation Model

  • Model: one scikit-learn random forest per recommendation pattern, 100 trees each
  • Training: the latest scan of every domain the analyzer measured in the last 90 days: 716 domains, 44 patterns, trained 2026-11-17
  • Held out of training: 144 domains, on which it is 94.9% accurate (macro F1 0.79)

4. Federated Threat Model

  • Model: logistic regression over 28 features of each request, scoring at POST /ai/federated-threat-model (how a round works)
  • Training: each proxy node trains on its own access log with DP-SGD (differential privacy) and sends only the change in weights; the API averages them (FedAvg): round 897, learned from 105,489 requests
  • Held out of training: each node scores its newest hour before training on it โ€” api3 3 of 3 hostile requests found, ROC AUC 1.00, mail 105 of 328 hostile requests found, ROC AUC 0.90, ns4 2 of 48 hostile requests found, ROC AUC 1.00 and web1 24 of 36 hostile requests found, ROC AUC 0.92 (a node's own refusals are its labels: where it refuses by address or rate, which a single request does not show, the model finds few)

5. Bot Classifier

  • Model: gradient-boosted trees (scikit-learn) over 22 features of each visitor-day in the proxy's access log: pages and the subresources a browser fetches with them, HTTP versions, status codes, pacing and the user agent's shape
  • Labels: the visitors the rules flag as automated, and the ones whose browser sent the analytics tracker's beacons: 16,917 and 800 over 8 days, trained 2026-11-17 18:00 UTC
  • Held out of training: 2026-11-16, ROC AUC 0.9996; at its threshold 97.3% of the automated visitors found, 0 of 33 browsers flagged
  • Found: 2,641 of the 23,426 visitor-days no rule or beacon labelled, left out of the analytics

โš™๏ธAutomation & Scheduled Jobs

Every scheduled job on the web server, each a systemd timer and a service with its own account and sandbox, as the units describe themselves (58 on this host; the other nodes run their own share, such as the federated model's training):

Job When Runs
Backup: this server's databases dumped for the mirror every 6 hours remotellm-dump-dbs.sh
Backup: the backups checked (ages, sizes, the hosts still copied) every hour backup-health-check.sh
Backup: the backups checked deeply (borg check --verify-data) Sundays at 04:00 backup-health-check.sh
Backup: the mail host mirrored to fortress through this server every 15 minutes mail-to-remotellm.sh
Backup: api3's pentest results for /pentests/ every 6 hours pentest-sync.sh
Backup: api3 mirrored to fortress through this server every 15 minutes pentest-to-remotellm.sh
Backup: the day's archive of the mirror on fortress daily at 06:50 remotellm-sync.sh
Backup: this server mirrored to fortress every 15 minutes remotellm-sync.sh
Backup: fortress's own state, kept here in borg daily at 07:30 remotellm-self-backup.sh
The proxy's blocklists from the database to its files (bots, fingerprints, countries) every minute sync_blocklist.sh
HTTP/3 conformance: the weekly matrix (four runs, promoted when every gate passes) Sundays at 04:20 weekly-matrix.sh
HTTP/3 conformance: a matrix of the running proxy build, when the published one measured another every hour matrix-if-new-build.sh
HTTP/3 conformance: the published matrix's provenance checked (alert raised or cleared) daily at 05:40 alert-provenance.sh
Cryptographic asset discovery: this host's keys, certificates, code, databases and services daily at 02:00 pqcrypta-discovery
Health of the self-hosted authoritative DNS (every name server and the signer) every 5 minutes pqcrypta-dns-health
Traffic, events and configuration of the self-hosted authoritative DNS (every name server and the signer) every 5 minutes pqcrypta-dns-stats
ECH: every node's new ECHConfigList in DNS, once each opened (web1) weekly pqcrypta-ech-publish
The proxy's WAF and rate-limit counters, added to the totals /security-systems/ shows every 5 minutes pcre.jit=0
The federated threat model: trained on web1's traffic, the update sent to the API every hour federated_node.py
The JA4 directory: the fleet's corpora collected and published for /ja4/ every 10 minutes sync-ja4-directory.php
The fleet's data files refreshed and distributed (JA3/JA4, GeoIP), and the proxy configs' drift checked Sundays at 03:15 refresh_fingerprint_db.sh
The other nodes' failed jobs collected for web1's alerts every 5 minutes pqcrypta-fleet-units
Keep the DNS names and the WAF whitelist entry for fortress on its current address every 5 minutes pqcrypta-fortress-address
The platform's health check: every engine's key generation, encryption and decryption, and every endpoint every 15 minutes pqcrypta-health-check
Scheduled jobs: the failure alerts of units that have since run clean, resolved every 5 minutes pqcrypta-job-alert
Analytics: the platform's diagnostics (connectivity, gaps, personal data, slow queries) Mondays at 02:10 analytics_diagnostics.sh
The proxy's blocklist: expired entries dropped every hour psql
Data retention: operational data past its retention purged (once enforcement is on) daily at 03:40 psql
Paths root runs code from that another account can change daily at 05:50 pqcrypta-privilege-audit
The proxy's config, redacted, published for /proxy-plane/ every hour sanitize-proxy-config.php
pqcrypta-proxy hang watchdog (restarts the proxy if it stops answering) every 30 seconds pqcrypta-proxy-watchdog.sh
HTTP/3 scanner: the Sunday follow-up (Claude Code on a branch of its own, for a human to review) Sundays at 06:30 sunday_followup.sh
Analytics: consented and anonymous counts reconciled daily at 01:10 reconcile_analytics.php
Bans: repeat offenders' bans lengthened from the evidence against them every hour risk_based_expiration.php
Analytics: automated visitors found by a model trained on the site's own traffic every 6 hours bot_classifier.py
Analytics: bot and scanner traffic removed from the page views daily at 02:00 cleanup_bot_analytics.sh
Bots found in the proxy's access log: probes, the honeypot feed, blocks every 5 minutes detect_bots_from_logs.sh
web1's Apache confs and config.php checked against pqcrypta-configs Sundays at 03:20 pqcrypta-config-drift
HTTP/3 scanner: every domain of the directory scanned again (the weekly sweep) Sundays at 04:17 'exec
IndexNow: the sites' new and changed URLs submitted every hour indexnow-submit.sh
The news page's articles refreshed from newsdata.io (through the API) every 6 hours refresh_news.sh
PQC scanner: every domain of the directory scanned again (the weekly sweep) Sundays at 02:17 'exec
PQC scanner: the domains the running scanner has not measured yet every hour 'exec
Analytics: bots found in the proxy's access log every hour parse_proxy_log_bots.php
Analytics: the proxy's access log into the anonymous page views and daily totals every 15 minutes parse_proxy_log_views.php
The Wizard's document index rebuilt daily at 02:00 run_rag_indexer.sh
HTTP/3 scanner: what the Sunday sweep did, reported Sundays at 06:05 'exec
Threat intelligence: the public feeds into threat_intelligence every 6 hours fetch-threat-intel.php
Upstream reports: where each fix stands, from the projects' own trackers every 3 hours upstream-status.php
pqpdf: the MITRE ATT&CK technique table refreshed from MITRE's release monthly on the 1st at 04:40 update_mitre_attack.py
pqpdf: the scanner's ML models retrained on the scans so far every 30 minutes train.py
pqpdf threat intelligence: the bazaar feed downloaded into the database daily at 02:30 update_ti_feeds.py
pqpdf threat intelligence: the feodo feed downloaded into the database daily at 02:15 update_ti_feeds.py
pqpdf threat intelligence: the hashes feed downloaded into the database Sundays at 03:00 update_ti_feeds.py
pqpdf threat intelligence: the openphish feed downloaded into the database every 6 hours update_ti_feeds.py
pqpdf threat intelligence: the threatfox feed downloaded into the database daily at 02:00 update_ti_feeds.py
pqpdf threat intelligence: the urls feed downloaded into the database every 30 minutes update_ti_feeds.py
pqpdf's temporary directories left in /tmp, removed after an hour every 30 minutes pqpdf-tmpclean

๐Ÿ”Security Infrastructure

1. TLS

As the Post-Quantum Web Analyzer last measured pqcrypta.com (2026-11-17 18:28 UTC), one group and one suite offered at a time:

  • Protocol: TLS 1.3 only โ€” no older version answered (the proxy's min_version = "1.3")
  • TLS over TCP: key exchange in the server's order MLKEM1024, SecP384r1MLKEM1024, X25519MLKEM768, SecP256r1MLKEM768, curveSM2MLKEM768, MLKEM768, frodo1344shake, hqc5, frodo976shake, hqc3, secp384r1, secp256r1, x25519; cipher suites TLS_AES_256_GCM_SHA384, TLS_CHACHA20_POLY1305_SHA256
  • QUIC v1 (RFC 9000): key exchange in the server's order MLKEM1024, SecP384r1MLKEM1024, X25519MLKEM768, SecP256r1MLKEM768, curveSM2MLKEM768, MLKEM768, frodo1344shake, hqc5, frodo976shake, hqc3, secp384r1, secp256r1, x25519; cipher suites TLS_AES_256_GCM_SHA384, TLS_CHACHA20_POLY1305_SHA256
  • QUIC v2 (RFC 9369): key exchange in the server's order MLKEM1024, SecP384r1MLKEM1024, X25519MLKEM768, SecP256r1MLKEM768, curveSM2MLKEM768, MLKEM768, frodo1344shake, hqc5, frodo976shake, hqc3, secp384r1, secp256r1, x25519; cipher suites TLS_AES_256_GCM_SHA384, TLS_CHACHA20_POLY1305_SHA256

HSTS: max-age=63072000; includeSubDomains; preload

2. Content Security Policy (CSP)

The policy this page was sent with, as every page sends its own (a nonce per request in place of {nonce}):

default-src 'self';
script-src 'self' 'nonce-{nonce}' 'nonce-{nonce}';
style-src 'self';
img-src 'self' data:;
font-src 'self' data:;
media-src 'self';
connect-src 'self' https://api.pqcrypta.com;
worker-src 'none';
child-src 'none';
object-src 'none';
script-src-elem 'self' 'nonce-{nonce}' 'nonce-{nonce}';
upgrade-insecure-requests;

3. Web Application Firewall (WAF)

  • Rules: the proxy's own 256 scored rules on every transport, HTTP/3 included, then the API's WAF on every API request (how they work)
  • Rules by kind: 58 bot, 54 scanner-probe, 18 SQLi, 18 XSS, 16 SSRF, 16 traversal, 15 command injection, 12 deserialization, 9 SSTI, 8 anomaly, 7 JNDI, 6 LFI, 6 NoSQLi, 5 XXE, 3 CRLF, 3 prototype pollution, 2 GraphQL; each scored by severity, the request refused past a threshold
  • Rate Limiting: per API key, signed-in user, TLS fingerprint, client address (as a proxy in front forwards it) and connecting address, the first a request carries, in the proxy (the limits themselves are left out of its published configuration); and per API key in the API
  • IP Blocking: from the proxy's access log every 5 minutes, with repeat offenders' bans lengthened every hour from the evidence against them

4. Bot Protection

  • Verified Crawlers: a search engine's crawler is let through its burst only when its address reverse-resolves to the engine's domain and that name resolves back to the address, as each engine documents; a User-Agent alone proves nothing
  • Human Verification: a proof-of-work challenge (ALTCHA's format) the browser solves, signed so nothing is stored when it is issued, harder for a caller whose last verification looked automated; no third-party CAPTCHA
  • Honeypots: 383 decoy paths no visitor has reason to ask for; a request to one is evidence against its sender

5. Privacy

  • Without consent: anonymous page views read from the proxy's log, with a visitor hash whose salt changes every day
  • Address masking: a record kept past its period loses its address and place
  • Retention: per table, purged daily at 03:40; operational data at most 90 days (the privacy policy)
  • Your Rights: access, correction, deletion and portability, on request (the privacy policy says how)

๐Ÿ“ŠPerformance Metrics

API Performance

  • Response Time: 84 ms median, 592 ms at the 95th percentile, over the 691 endpoints of the last health run (each a real request, key generation and encryption included)
  • Uptime: 99.96% of the last 30 days' health runs found the API answering; 99.77% of 1,952,240 endpoint checks passed
  • Throughput: the proxy served up to 72,200 requests a second (h3, 1k body, 10 connections; the median of 6 runs on AMD EPYC-Milan Processor, 2026-09-25), HAProxy 66,700 on the same machine; every cell is on the proxy comparison

Algorithm Performance

Key generation, encryption and decryption through the API, the median of 5 full benchmark runs (the full distribution):

  • Classical: 16 ms key generation, 16 ms encryption, 15 ms decryption
  • Hybrid: 16 ms key generation, 22 ms encryption, 17 ms decryption
  • Post-Quantum: 14 ms key generation, 17 ms encryption, 16 ms decryption
  • Post-ZK Homomorphic: 2.12 s key generation, 536 ms encryption, 901 ms decryption

Database Performance

  • Connection Pool (pages): PgBouncer 1.26.1, pooled by transaction, a backend closed after 60 s idle, on a local socket: a page holds a server backend only while its statement or transaction runs
  • Connection Pool (API): deadpool-postgres, inside the API, kept between its minimum and maximum (the API's DATABASE_* settings)
  • Statement Time: 2.2 ms on average over 133,651,901 statements since 2026-10-18 (pg_stat_statements)
  • Size: 28 GB

Frontend Performance

Core Web Vitals as visitors' browsers measured them over the last 28 days: the 75th percentile of 291 page loads, as Web Vitals are judged.

  • Largest Contentful Paint (LCP): 918 ms
  • Interaction to Next Paint (INP): 142 ms
  • Cumulative Layout Shift (CLS): 0.000
  • First Contentful Paint (FCP): 826 ms
  • Time to First Byte (TTFB): 590 ms

๐Ÿ› ๏ธTechnology Stack

Frontend

  • PHP: 8.4.27
  • JavaScript: ES2024
  • Build Tool: Vite 8.3.7
  • Libraries: @noble/post-quantum 0.7.1, @noble/curves 2.4.0, Three.js 0.187.0, Chart.js 4.4.1, D3.js 7.8.5

Backend

  • Language: Rust 1.99.0, edition 2024
  • Framework: Axum 0.8.9 (async web framework)
  • Runtime: Tokio (async runtime)
  • Database Driver: tokio-postgres 0.7.18, pooled by deadpool-postgres 0.14.2

Database

  • PostgreSQL: 16.15
  • Connection Pool (pages): PgBouncer 1.26.1, pooled by transaction, a backend closed after 60 s idle
  • Connection Pool (API): deadpool-postgres, inside the API
  • Backup: BorgBackup archives and database dumps, on systemd timers, copied off the host

Reverse Proxy

  • pqcrypta-proxy: Rust-based reverse proxy
  • Protocols: HTTP/3, QUIC, HTTP/2, WebTransport
  • TLS: rustls on QUIC, OpenSSL 4.0.3 on TCP; ML-KEM native in both
  • WebTransport: over HTTP/3, UDP 443

ML/AI

  • Python: 3.12.3
  • Libraries: scikit-learn 1.7.1 (the HTTP/3 recommender), NumPy 2.2.6 (the federated model's DP-SGD), Qiskit 2.5.2 (the quantum models)
  • Embeddings: sentence-transformers
  • Vector Store: FAISS
  • LLM API: openai/gpt-oss-120b on Groq

๐ŸŽฏSystem Status

As of the last health run: 2026-11-17T18:18:01Z

Operational Status

  • โœ… 35 of 35 Algorithms: key generation, encryption and decryption all passed in the last health run
  • โœ… 691 API Endpoints: Monitored every 15 minutes with real cryptographic tests โ€” 690 of 691 healthy at the last run
  • โœ… 96 Pages: every one requested through PHP-FPM by the release tool before a release goes live
  • โœ… ML/AI Services: the federated threat model (round 897) and the HTTP/3 recommendation model (trained 2026-11-17) live; the Wizard and its RAG index off
  • โœ… Database: PostgreSQL 16.15, retention enforced table by table
  • โœ… Security: WAF active, JA3/JA4 fingerprinting, bot protection, TLS 1.3 with PQC hybrid across 5 proxy nodes

Recent Changes

The pull requests this release brings and the ones before it, newest first:

  1. 2026-11-17 (#531) โ€” pqcrypta-proxy pin: afdd107, v0.16.0 (chrono and humantime replaced with jiff, HTTP dates in all three RFC 9110 forms)
  2. 2026-11-17 (#530) โ€” /upstream-bugs/: maintainers' reviews of pull requests; noq#879's next step
  3. 2026-11-16 (#529) โ€” /what/: the pages' connection pool, PgBouncer, beside the API's
  4. 2026-11-16 (#528) โ€” pqcrypta-configs pin: 30f47c9, PgBouncer on web1 (configs #164)
  5. 2026-11-16 (#527) โ€” Migration 136: the pgbouncer schema belongs to the role applying it
  6. 2026-11-16 (#526) โ€” HTTP/3 timeline in one pass: ranks from an index, comparisons by lag(), rows by key
  7. 2026-11-16 (#525) โ€” pqcrypta-proxy pin: 0.15.0 (f6a133a), the release deployed to all five nodes; /pqcproxy/ install as the release ships it
  8. 2026-11-16 (#524) โ€” Menu: the proxy matrix's figures from the matrix, and the benchmark entry as the page now stands
  9. 2026-11-16 (#523) โ€” pqcrypta-proxy pin: 0.14.1 (20cd151), the release deployed to all five nodes
  10. 2026-11-16 (#522) โ€” pqcrypta-proxy pin: 0.14.0 (47b915a), the release deployed to all five nodes
  11. 2026-11-16 (#515) โ€” /upstream-bugs/: our own reports name their fix too
  12. 2026-11-16 (#514) โ€” /security-systems/: the proxy counters' window in minutes while it is short

โŸจ QUANTUM ERROR PORTAL โŸฉ

Navigate the Error Dimensions