Nyx Multi-Dialect Architecture Multi-Dialect Unified Trunk • 4 Memory Models (1 Main + 3 Companion Dialects) • One Sovereign Toolchain
View Architecture Roadmap →
Sovereign Language Extensions

Beyond Single-Paradigm Limits:
The Nyx Multi-Dialect Architecture

Rather than forcing one rigid memory model on every domain, Nyx extends into 3 specialized companion dialects (Zephaniah, Havilah, Jude) on top of the Nyx Base engine, sharing the exact same compiler trunk, MLIR/LLVM codegen, standard library, and African language aliases (Hausa, Yorùbá, Igbo).

One Unified Toolchain: nyxc Zero-Cost Interop: Direct ABI shims Sovereign Heritage: 9jaonCloud Research

Multi-Dialect Architectural Hierarchy

All companion dialects compile through the shared Nyx driver and lower into identical MLIR dialect representations, allowing zero-cost cross-dialect calling conventions. 

graph TD
    subgraph Base [Nyx Unified Foundation]
        NYX[🌙 Nyx Base Engine & OS
Region Arenas + Zero-GC]
    end
    
    subgraph Dialects [3 Companion Dialects]
        ZEPH[⚡ Zephaniah .zeph
Gradual Ownership & ARC
Flag: --mode=gradual]
        HAV[🛡️ Havilah .hav
Linear Capability = Ownership
Flag: --mode=capability]
        JUDE[📐 Jude .jude
SMT Formal Proofs & Z3
Flag: --verify]
    end

    subgraph Cybernetics [Sovereign Cybernetic Substrates]
        SERAPH[🤖 SERAPHIM
1 kHz Whole-Body Humanoid Robotics]
        SOPHIA[🧠 Nyx SOPHIA
7-Layer AGI & World Model]
    end

    NYX --> ZEPH
    NYX --> HAV
    NYX --> JUDE
    
    ZEPH --> SERAPH
    HAV --> SERAPH
    JUDE --> SERAPH
    SERAPH --> SOPHIA

The 3 Companion Extension Dialects

Select any companion dialect below to explore its syntax, formal semantics, compiler flags, and industrial use cases:

⚡
.zeph • --mode=gradual

Zephaniah

Ergonomics & Gradual Ownership: Designed for rapid development and application code where strict lifetime annotations are burdensome. Uses progressive strictness and automatic ARC fallback.

  • Zero Lifetime Annotations: Write high-level script-like code with C performance.
  • Gradual Borrow Checker: Enforces borrow checking only where declared.
  • Ideal For: Web services, UI controllers, scripts, and high-level business logic.
Explore Zephaniah →
🛡️
.hav • --mode=capability

Havilah

Linear Capability = Ownership: Minimalist architectural security where resources are linear capability tokens. Dropping the capability immediately and deterministically frees the underlying resource.

  • Linear Type System: Capabilities cannot be duplicated, leaked, or forged.
  • Drop = Deallocate: No garbage collection and no region tracking overhead.
  • Ideal For: Microkernel drivers, hardware peripheral sandboxes, multi-tenant cloud.
Explore Havilah →
📐
.jude • --verify

Jude

SMT Formal Mathematical Verification: Embeds first-order logic pre/post condition contracts (@verify) directly into systems code, discharging proof obligations to Z3/CVC5 solvers.

  • Machine-Checked Proofs: Zero runtime assertion cost; verified at compile-time.
  • Cryptographic Certificates: Emits standalone .jude.cert audit artifacts.
  • Ideal For: Avionics flight control, medical devices, cryptography, defense systems.
Explore Jude →

Sovereign Cybernetic Substrates (Robotics & AGI)

Built on the multi-dialect safety and mathematical guarantees of Nyx, these two high-level platforms power real-world robotics and embodied intelligence:

🤖
Project SERAPHIM • Private Flagship

SERAPHIM

Real-Time Humanoid Robotics Substrate: End-to-end physics, rigid-body Featherstone \(O(n)\) dynamics, Whole-Body Control (WBC) Quadratic Programming, and 1,000 Hz zero-GC motor control loops.

  • Zero GC Pause: Hard deterministic sub-0.5ms loop jitter.
  • Rigid Body Physics: Spatial 6D Plücker vector algebra.
Explore SERAPHIM Robotics →
🧠
Project SOPHIA • Private Flagship

Nyx SOPHIA

7-Layer Cognitive AGI Architecture: DreamerV3-style RSSM latent world model, episodic/semantic HNSW memory substrate, 3-tier goal hierarchy, and epistemic self-reflection.

  • Latent Imagination: Counterfactual reasoning in latent space without physical risk.
  • Embodied Action: Direct integration with SERAPHIM WBC motors.
Explore Nyx SOPHIA AGI →

4 Memory Models on 1 Unified Compiler Trunk

Nyx eliminates language fragmentation by providing 4 distinct memory & ownership models (1 Main Base + 3 Companion Dialects) on a single compiler and OS substrate, plus two applied cybernetic platforms:

Dialect / Category Extension / Flag Memory & Safety Model Primary Industrial Use Case Verification Mechanism
1. Nyx Base (Main) .nyx Region Bump Arenas + ARC Fallback Systems software, OS kernels, cloud web engines Compile-Time Escape Analysis
2. Zephaniah (Dialect) .zeph (--mode=gradual) Gradual Ownership & Transparent ARC Rapid application engineering, high-level business logic Gradual Borrow Checking ([Z003])
3. Havilah (Dialect) .hav (--mode=capability) Linear Capability Tokens (Cap = Ownership) Microkernel sandboxes, hardware device drivers Linearity Single-Use Enforcement ([H001])
4. Jude (Dialect) .jude (--verify) SMT Contract Proof Obligations Avionics, medical devices, defense, cryptography Z3/CVC5 Machine-Checked SMT Solvers
SERAPHIM (Project) projects/seraphim Lockless Real-Time 1 kHz Memory Pools Humanoid locomotion, bipedal balance, robotic arms Whole-Body QP Feasibility & Watchdogs
Nyx SOPHIA (Project) projects/sophia Multi-Modal Dual-Trace Substrates Autonomous embodied agents, AGI cognitive planning Epistemic Uncertainty Estimation (MSE)
Industry Language Architecture Comparison:
While single-model languages (such as Rust with strict lifetime annotations, Zig with manual allocators, Vale with generational references, and Toka with handle-soul sigils) force all code into a single paradigm, Nyx gives engineers the sovereignty to select the optimal model per file (.nyx, .zeph, .hav, .jude) with 100% ABI and zero-cost interop.

The Nyx Multi-Dialect Sovereign Architecture is developed by Simeon Bala • 9jaonCloud Sovereign Systems Research.

nyx@9jaoncloud.com.ng LLM Provenance Policy Research Papers Directory