sovereign-llm — Quantum-Seeded Sovereign LLM Engine

Rust binary wrapping Mistral GGUF inference with genuine quantum entropy.
Built by Ahmad Ali Parr · SnapKitty Collective

Pipeline

ANU QRNG (Australian National University quantum vacuum)
    ↓ 32 bytes → Born-rule collapse
    → unique temperature per run [0.15, 1.2]
    → HKDF-SHA256 seed derivation
snapkitty-mistral (Ollama / llama.cpp CUDA)
    ↓ bool formula (NAND attention) + softmax over 32k vocab
SLC membrane (5 regex banks, 3-pass recursive)
    ↓ pass / sanitize / quarantine / reject
WORM chain (SHA-256 append-only)
    ↓ QUANTUM_ENTROPY seal + GENERATION seal
Sovereign answer

The Math

An LLM is:

1. Bool formula — NAND decomposition of transformer attention (proven in DOI 10.5281/zenodo.21351461)
2. Softmax — picks the next token from the logit distribution

We replace the PRNG sampler with actual quantum vacuum fluctuations from ANU.
Every generation starts in a genuinely different quantum state.

Usage

sovereign-llm --prompt "your query" --agent BOB

Trust

Bel Esprit D'Accord Irrevocable Trust · EIN 42-6976431