neuralspace-ai 1.0.0

Cognitive Code Topology Engine - Self-evolving neural security scanner

🧠 NeuralSpace – Cognitive Code Security Engine

NeuralSpace is a next‑generation, self‑evolving security scanner that detects obfuscated malicious code using a Covalent Tree topology – a novel neural architecture that routes code through a fractal decision tree where each branch has its own specialized "neural atom."

Unlike traditional static scanners (SAST/NGAV) that rely on rigid rule sets or giant black‑box models, NeuralSpace learns the semantic fingerprint of your codebase and autonomously fractures into new branches when it encounters novel architectural patterns.

---

🔥 The Problem We Solve

Current Tool	Limitation	NeuralSpace Advantage
Traditional AV	Relies on known signatures.	Blocks zero‑day obfuscated threats.
SAST (SonarQube)	99.5% false positives.	Contextual detection (e.g., requests.get alone is safe; requests.get + exec is not).
Transformer Models	Huge, slow, cloud‑dependent.	Lightweight (~8 KB weights), runs instantly on CPU.
File Watchers	React to files, don't understand content.	Routes files dynamically into a living knowledge tree.

---

✨ Key Features

• 🧬 Self‑Evolving Topology – The Covalent Tree spawns new branches when it detects structural drift in your code (cosine similarity < 0.15). It doesn't just classify; it organizes your codebase.
• 🧠 Distributed Neural Atoms – Each tree branch has its own PureNeuralAtom (512→4→4 network) initialized with a unique seed. This creates specialized "brains" for different code families (e.g., web scrapers vs. quant math vs. CLI tools).
• 🛡️ Obfuscation‑Resistant Tokenizer – Engineered combination features (indices 490–495) catch multi‑stage evasions like base64 + exec + requests, bypassing the neural network's natural struggle with long‑range dependencies.
• 🌍 Polyglot – Scans Python, JavaScript, TypeScript, Go, and Rust (more coming soon).
• ⚡ Ultra‑Lightweight & Local – Trains in under 1 minute on a standard CPU. No cloud, no GPU, no expensive API calls.
• 🔗 CI/CD Native – Ships with a pre‑built GitHub Action that runs on every push and pull_request.
• 🌐 Federated Intelligence – Shares anonymized threat signatures with a global network, creating a living immune system.

---

🏗️ How It Works

1. Tokenization – Raw source code is converted into a 512‑dimension vector using unigram hashing, byte trigrams, and hard‑coded combo indices (requests+exec → +8.0).
2. Routing – The vector descends the Covalent Tree. If it matches a child node (cosine similarity > 0.85), it dives deeper. Otherwise, it stops.
3. Judgment – The terminal node's PureNeuralAtom computes two scores:
   • Sentinel (S) – Threat probability (class 3).
   • Logic (L) – Safe probability (class 0).
4. Enforcement – If S > 0.25 or L < 0.2, the file is quarantined (renamed/moved) and logged.
5. Evolution – If the file is allowed but deviates significantly from the parent's latent centroid, the tree spawns a new child node to capture this new architectural pattern.

---

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/krishnakanthsharmat-cloud/NeuralSpace.git
cd NeuralSpace

# Install in development mode (recommended)
pip install -e .