Alluci AI - The Sovereign Agent & Polytope Projection Network

Welcome to the central repository for the Alluci Sovereign Agent and the Polytope Model Family.

Alluci AI is dedicated to building local-first, zero-friction, sovereign intelligence systems. We believe that true artificial intelligence must be geometrically bounded, mathematically deterministic, and intrinsically tied to the biological state of the user, functioning entirely within a local cryptographic perimeter.

The Alluci Sovereign Agent Architecture

The Alluci-Sovereign Agent is a distributed, agentic runtime acting as your primary digital interface. By merging strategic hierarchical planning with a local-first execution gateway, it coordinates across a secure 20+ multi-bridge ecosystem. It is engineered for 100% data sovereignty, ensuring your personal and professional digital life remains cryptographically secured and physically under your control.

1. Executive Orchestration & DAG Planner

Hierarchical DAG Planner: Complex user objectives are parsed into a Directed Acyclic Graph (DAG). Sub-tasks are executed in parallel across multiple worker agents, dramatically reducing latency for multi-step goals.
Token Optimization: The Supervisor Agent condenses verbose worker outputs into dense "Sovereign Context Tokens" before passing them down the DAG, preventing context-window bloat.

2. Simplicial H-LSM Memory & The DPK

The Discrete Projection Kernel (DPK): Replaces slow floating-point database math with $\mathcal{O}(1)$ complexity integer-based lookups, providing sub-microsecond latency.
Topological Barcodes: Memory isn't retrieved via keyword search. The agent generates Betti Signatures and retrieves context based on "Structural Homeomorphism" (logical shape), guaranteeing absolute context relevance.

3. The "Dream" Cycle & Autonomous Evolution

When the system detects low cognitive load (e.g., when you are asleep), the daemon reallocates 100% of hardware resources to internal evolution:

Cognitive Distillation: Analyzes the day's interactions using Socratic questioning, distilling episodic logs into permanent Semantic Truths.
Native DPO & LoRA Forge: Utilizes a native PyTorch Direct Preference Optimization (DPO) kernel to forge custom LoRA adapters using your specific data overnight.

4. Hardware-Level Security & Verus Identity

The Sovereign Kill Switch: If the agent attempts a high-cognitive action (banking, crypto tx) and your Apple Watch is off-wrist or no pulse is detected, execution is instantly aborted.
Verus Node Integration: Full integration with the Verus blockchain to secure your identity (VerusID), manage sovereign funds, and sign transactions cryptographically within the vault.

The Alluci Polytope Family

The Alluci Polytope models are a fleet of highly specialized variants forged from the Google Gemma 4 family. They serve as the brain of the Sovereign Agent. Ranging from ultra-lightweight Edge nodes to massive local cognitive engines, these models have been fine-tuned using proprietary Structural Datasets to eliminate soft, probabilistic text-based behaviors.

Instead of text prompts, these variants operate on Topological Boundaries.

The Fleet

Our models are natively forged for both Apple Silicon (Unified Memory) and Windows/Linux Edge (CUDA/CPU):

Model Name	Variant/Size	Primary Function
Edge Canary	E2B	Instantaneous micro-routing and system telemetry
Local Standard	E4B & 12B	The primary reasoning layer for most developers
Heavy Compute	26B-A4B & 31B	Deep, offline massive cognitive processing

Formats Available:

MLX (.safetensors): Forged natively via mlx-lm for Apple Silicon.
GGUF: Compiled via llama.cpp for maximum efficiency on Windows/Linux.

Safety-By-Construction (The PPN)

The Alluci Sovereign Agent operates exclusively via the Polytope Projection Network (PPN).

Rather than relying on probabilistic guardrails or post-hoc system prompts to manage AI behavior, the PPN utilizes a deterministic geometric Polytope. This is a bounded, n-dimensional convex geometric region that defines the absolute boundary of safe, valid, and admissible states.

Vertices (Axiomatic Core States): The absolute, verified safe limits and fundamental system laws.
Simplices (The Structural Faces): The valid structural pathways the agent's logic can navigate without causing "Manifold Tearing."
Betti Signatures: Topological invariants ensuring that a path to the safe-halt condition is always reachable.
Half-Spaces: Using sub-microsecond GJK (Gilbert-Johnson-Keerthi) projections, any high-entropy output that drifts outside the boundary is instantly and deterministically snapped back onto the Polytope's shell.

The Affective Computing Engine (ACE)

The Alluci Sovereign Agent is biologically bound to you.

The Affective Computing Engine (ACE) creates a real-time, closed-loop system between your physiological state (via Apple Watch HealthKit streams) and the local AI environment. Using zero-latency fixed-point arithmetic via libaffect_kernel.dylib, the ACE maps your biometric data directly into the Polytope Manifold:

Arousal Matrix: Driven by Heart Rate and Respiratory Rate. Dilates the Polytope axis.
Tension: Driven by Heart Rate Variability (SDNN). High stress restricts the topological boundaries of the agent, forcing highly conservative, safe execution pathways to reduce your cognitive load.
Valence: Modulated by sleep efficiency and HRV stability. Positive valence unlocks lower-retention episodic memories for creative discovery.

System Behavior: When you are stressed, the AI becomes rigid and protective. When you are in a flow state, the AI expands its latent space for maximum creative bandwidth.

The Sovereign Agent Installer

To seamlessly run these models locally and tap into the architecture described above, download the Alluci Sovereign Agent One-Click Installer from our primary GitHub repository.

The embedded HardwareScanner will automatically detect your OS, Silicon type, and RAM overhead, and natively pull the perfect Polytope variant from this Hugging Face organization directly onto your machine.

No cloud. No APIs. Pure sovereign compute.