Tensor Reality: Tensor-Based Computational Engine

An advanced, high-performance computational engine written in Rust. It models physical reality as a resource-constrained, autoregressive tensor process designed for autonomous AI agents and complex environmental simulations.

🚀 Key Features

• Autoregressive Tensor Kernel: Step-by-step state propagation engine tracking multi-dimensional token interactions over discrete time slices (ticks).
• Attention Heatmaps & Visualization: Real-time terminal visualization of context fields, attention densities, and core execution statistics.
• State Persistence: Automated snapshot serialization into full JSON states and text-based operational checkpoints.
• Massive Parallelism: State propagation and spatial heatmap computations are accelerated using rayon for data-parallel operations.
• Resource-Constrained Mechanics: Conservation laws mapped to data structures via metrics like "Total Mass" and dynamic context length limits.

🛠️ Architecture & Core Modules

The engine is decoupled into specialized domains:

• kernel (TensorKernel) — Coordinates the execution pipeline, ticks, and context buffer mutations.
• math — Handles raw multi-dimensional matrix operations and tensor transformation utilities.
• storage — Controls serialization (save_full_state, load_full_state, save_checkpoint) powered by serde_json.
• viz — Terminal rendering systems (print_heatmap, print_stats, print_compact) for tracking simulator state.
• utils — Auxiliary helper routines and internal primitives.

📥 Installation & Setup

Ensure you have the Rust toolchain installed.

git clone https://github.com/Rachkovskiy-I/tensor_reality
cd tensor_reality
cargo build --release

💻 Usage & Lifecycle

The engine automatically handles hot-reloads based on current execution snapshots:

1. State Recovery: Attempts to locate and deserialize full_state.json.
2. Context Hydration: Restores previous tokens, current tick metrics, and re-allocates core tensor structures.
3. Execution Loop: Executes targeted step computations (kernel.step()) over 100 or 1000 iteration epochs.
4. Analysis & Flush: Outputs terminal attention heatmaps, computes current context entropy, and serializes state back to disk.

cargo run --release

🤖 LLM & AI Agent Ingestion Context

• Context Dynamics: Tokens inside the context carry specific physical weights and properties ("total_mass").
• State Operations: Modifying the tensor mechanics should preserve conservation invariants evaluated inside tests/integration_tests.rs.