Internal Neural Network Security Engine & Hybrid AI Health Advisor

Docs / StarVault guide / Internal Neural Network Security Engine & Hybrid AI Health Advisor

100% Offline & Zero-Knowledge by Default: Analyzes multi-dimensional credential telemetry entirely in memory using a dedicated local 2-layer perceptron (\(12 \to 16 \to 8\)) with LeakyReLU activations without transmitting a single byte to external servers.
Sub-Millisecond Inference (< 0.5ms): Pure vectorized matrix arithmetic evaluates your entire vault in < 1ms with zero heavy GPU/PyTorch dependencies.
12-Dimensional Security Vector Analysis:
- length_norm, entropy_norm, char_variety, pattern_cleanliness, age_stale_factor, expiry_urgency, exposure_surface, mfa_protection, reuse_count_norm, group_breadth, metadata_completeness, key_algorithm_strength.
Tri-Head Neural Outputs:
- Compromise Risk Score (0-100): Sigmoid-activated probability of credential vulnerability.
- Anomaly Score (0-100): Detects anomalous configurations (e.g. unencrypted HTTP with admin privileges and missing 2FA).
- 6-Class Softmax Classification: SECURE_OPTIMAL, CRITICAL_EXPIRING, WEAK_ENTROPY, REUSED_CREDENTIAL, STALE_UNROTATED, UNPROTECTED_EXPOSURE.
Hybrid AI Architecture:
- Internal Neural Network is the default, primary security advisor.
- Optional cloud/local LLM integrations (OpenAI, Anthropic Claude, Google Gemini, Ollama, OpenRouter) can be enabled on demand for natural language prose advice and architectural suggestions.
1-Click Neural Scan & Auto-Fix: Run neural health audits in 1-click and immediately auto-rotate weak passwords, renew expiring tokens, or auto-assign missing groups and services.
Full Ecosystem Parity: Exposed seamlessly across Desktop GUI, Chromium Extension (/api/neural_health, /api/neural_analyze_entry), and Model Context Protocol (MCP) server.