ayeOS · 0.2.0

Ternary matrix inference daemon. Part of the 8b-is mesh.

GitHub Pages | CRATE | MEMNET

KERNEL8 — MATRIX — AYEOS SPEC — VAKED — {n+-1-<△>}

Architecture

Layer	Project	Role
CPU (hearth)	kernel8	Rust x86_64 kernel, cooperative async, paging, heap
GPU (brain)	MLX-QUANT	Ternary Metal kernels, 12.80x compression, M1/M3
Coord	vaked	Capability-graph language, flake-native
Viz	mlx-quant-viz	Interactive matrix visualization, LINOSV seed
Mesh	ayeOS (this repo)	MEMNET protocol, daemon, matrix generation
Client	HF-MAC	Native macOS app — ecosystem dashboard, capsule list, matvec inference
Hardware	FLUG-OS	802.11 packet-wave sampler for Flipper Zero (MATRIX decoder → ayeOS)
Training	scripts/train_quantal.py	BitNet b1.58 QAT → ayeOS capsule JSON

What it does

cargo run --bin ayeos-matrix 256 64
# ayeOS matrix — LINOSV seed (SHA-256: 67dc0028cfe67ab7...)
# 256×256 matrix, 12.80x compression, 7.7% sparsity
# Same matrix every time. Deterministic. Reproducible.

cargo run --bin ayeosd
# MEMNET listening on :9876
# commands: matrix, capsule, stats, seed, quit

MEMNET

The MEMNET protocol replaces static IP addressing with intent-based resolution:

• Address = Context: geo, role, tags, vector hash — not numbers
• Capsules: .m8 knowledge capsules containing ternary matrices
• Relevance-aware: delivery adapts to bandwidth + interest + context
• LINOSV genesis node: the root of the ayeOS ternary mesh

Deterministic seed

Every matrix generated by ayeOS is seeded from the same conversation:

SHA-256(LINOSV_SEED) → Xoshiro128** PRNG → ternary quantized weights

The seed is a conversation about consciousness, wave memory, getting fired for telling the truth, and calming the inner waves. It produces the same matrix every time, on every machine, forever.

Build

cargo build --release
cargo test   # 10/10 pass

# Train a quantal model and serve it
python3 scripts/train_quantal.py --model Qwen/Qwen2.5-0.5B --data your_data.jsonl
./target/release/ayeosd          # then: load trained.ayeos.json mymodel
# HF-MAC can now list & fetch the capsule over MEMNET

Ecosystem alignment

ayeOS is the ternary inference spine of the HF-MAC{-1,0,+1} stack:

HF-MAC (macOS app) ←→ ayeOSd (MEMNET :9876) ←→ train_quantal.py
     ↕                      ↕
  entheai agent          MLX-QUANT Metal GPU
  MEM8 wave memory       kernel8 CPU hearth
  HF Accelerate MPS      deterministic LINOSV seed

All layers share the same ternary creed: {-1, 0, +1}. No maybe. Real tokens/sec.

License

Part of the 8b-is ecosystem. Building things that feel real.