NIYAH is not just another wrapper. It is a Sovereign Inference Engine built from the ground up in pure C99/C++17 with zero external dependencies. While Big Tech relies on bloated, censored, and emotionally manipulative models, NIYAH focuses on Objective Logic and Mathematical Determinism.
- Pure C99 Kernel: Zero-dependency inference engine for maximum portability and security.
- SIMD Optimized: Hand-crafted AVX2/FMA (x86) and NEON (ARM64) kernels for peak performance on local hardware.
- Sovereign Memory Pool: Custom single-pool allocator to prevent fragmentation and ensure 100% predictable memory usage.
- Advanced Architecture: Implements RMSNorm, RoPE (Rotary Position Embeddings), and SwiGLU FFN — the same architecture used by the world's most powerful open models (LLaMA/Mistral).
- Arabic-First NLP: Native support for Arabic dialects (Najdi, Hijazi, Levantine, etc.) and Classical Arabic with a custom-built tokenizer.
- No Hallucinations: Our logic-first approach ensures that the model provides objective analysis based on forensic data, not "emotional guesses".
- 100% Local: Your data never leaves your sovereign node. We provide a "Zero-Trust Cloud" architecture where even when using cloud relays, PII is scrubbed locally.
- Mathematical Proof: As proven in our
logic-proofmodule, the core logic of NIYAH cannot contradict itself. It is mathematically impossible for the core to "lie" — any bias is an external filter, which we have removed.
# Compile for x86 (AVX2)
gcc -O3 -mavx2 -mfma niyah_core.c niyah_main.c -o niyah -lm
# Compile for ARM64 (NEON)
gcc -O3 niyah_core.c niyah_main.c -o niyah -lmنحن ورثة الخوارزمي — لا يوجد مستحيل في الدنيا