Nomos (Νόμος) CLI

Client-Side Hardware-Accelerated Decision Model Training & SafeTensors Exporter
Web Studio: https://nomos.dforge.ca

---

🔒 Open Source & Privacy Guarantee

Nomos is distributed as an open-source tool so engineering teams and security auditors can independently verify:

1. Zero Data Exfiltration: Training datasets and raw training examples never leave your machine. All gradient steps, loss calculations, and matrix operations run 100% locally on your workstation or server.
2. No Proprietary Lock-In: Calibrated models export into standard, zero-copy model.safetensors compatible with Hugging Face, Candle, ONNX Runtime, and PyTorch.
3. Auditable Codebase: No telemetry is dispatched without explicit user authentication, and no customer model weights are hosted on our infrastructure.

---

🚀 Key Capabilities

• Hardware-Accelerated BYOC (Bring Your Own Compute):
  • Apple Silicon Metal (MPS): Utilizes unified high-bandwidth memory on M-series chips.
  • NVIDIA CUDA: Tensor Core accelerated forward passes (Compute Capability 7.0+).
  • Vectorized CPU SIMD: AVX2, AVX-512, and ARM NEON fallbacks.
• Multi-Head Decision Architectures:
  • Calibrate up to 20 parallel probabilistic heads per forward pass (binary Bernoulli, ordinal score, and categorical choice).
  • Replace sluggish multi-second LLM agent roundtrips with sub-50ms deterministic neural evaluations.
• Turnkey Self-Hosting:
  • Exports a clean model.safetensors alongside schema.json and a lightweight Dockerfile ready for Kubernetes, AWS ECS, or bare-metal edge deployments.

---

📦 Installation

Pre-Built Binaries

Download the latest pre-compiled binary for your architecture from GitHub Releases:

# macOS (Apple Silicon / ARM64)
curl -L https://github.com/digitalforgeca/nomos-cli/releases/latest/download/nomos-macos-arm64 -o /usr/local/bin/nomos
chmod +x /usr/local/bin/nomos

# Linux (x86_64)
curl -L https://github.com/digitalforgeca/nomos-cli/releases/latest/download/nomos-linux-amd64 -o /usr/local/bin/nomos
chmod +x /usr/local/bin/nomos

Build From Source (Cargo)

Requires Rust 1.75+:

git clone https://github.com/digitalforgeca/nomos-cli.git
cd nomos-cli
cargo build --release
cp target/release/nomos /usr/local/bin/

---

⚙️ Quickstart & CLI Commands

1. Profile Your Workstation

Detect your hardware accelerators (Metal, CUDA, CPU SIMD) and view recommended batch sizes:

nomos profile

2. Authenticate (Optional)

Link your machine to your Digital Forge workspace using your API key (fk_live_...). Configuration and credentials are stored strictly in ~/.nomos/config.json:

nomos auth login --key fk_live_YOUR_KEY
nomos auth status

3. Discover Foundation Decision Backbones

Inspect available pre-trained decision models:

nomos catalog

4. Synthesize a 20-Head Schema From Natural Language

Transform a plain-English intent or engineering goal into a structured multi-head question schema:

nomos synthesize --goal "Triage incoming security alerts, filter noise, and route high-severity CVEs"

5. Train On-Premise

Train 100% locally with hardware acceleration:

nomos train \
  --dataset ./data/train.jsonl \
  --backbone modernbert-base \
  --epochs 3 \
  --batch-size 32 \
  --output ./output/my-decision-model

6. Evaluate Text

Run sub-50ms multi-head inference:

nomos evaluate --model my-decision-model --text "Unauthenticated remote code execution vulnerability discovered in ingress controller"

7. Export for Self-Hosting

Export zero-copy weights, decision schema, and a production Dockerfile:

nomos export \
  --model-dir ./output/my-decision-model \
  --output ./dist/production-bundle

---

📁 Configuration Directory

Nomos persists local configuration in:

~/.nomos/
└── config.json

Environment variables can override stored defaults:

• NOMOS_ENDPOINT: Remote studio API gateway (defaults to https://api.nomos.dforge.ca)
• NOMOS_API_KEY: API authentication key (fk_live_...)
• NOMOS_ORG_ID: Active workspace organization ID

---

📄 License

Licensed under the Apache License, Version 2.0 (the "License"). You may obtain a copy of the License in the LICENSE file or at:

http://www.apache.org/licenses/LICENSE-2.0