← Back to Platform

Multinex Shield

Zero-Trust AI Perimeter

Overview

Multinex Shield is the ultimate safeguard for enterprise intellectual property. It operates as an impenetrable, high-throughput bidirectional firewall sitting squarely between your internal networks and external LLMs. By weaponizing an ultra-low-latency Aho-Corasick algorithm, Shield intercepts, parses, and redacts restricted PII and proprietary logic in milliseconds—guaranteeing your corporate data never leaks into public training datasets.

Multinex Shield Topology

[ IDE / Agent Client ] ──(REST/WebSocket)──> [ Multinex Gateway (Deno) ]
                                                    │
                                                    ├──> 1. Rate Limiting / Auth
                                                    ├──> 2. [ Rust Classifier Node ]
                                                    │       (Aho-Corasick Pattern Match)
                                                    │       - PII Redaction
                                                    │       - Secret Stripping
                                                    │       - Prompt Injection Block
                                                    │
[ Public LLM (OpenAI/Vertex) ] <──(Sanitized)───────┘
                                                    │
[ Grafana / Prometheus ] <──(Telemetry)─────────────┘

The Execution Path

  1. Ingress: A client sends an OpenAI-formatted request to the local Deno Gateway.
  2. Classification: The gateway passes the payload memory to the Rust-based classifier, scanning for millions of prohibited patterns in O(n) time complexity.
  3. Mutation/Rejection: Hard-block rules trigger a 403. Redaction rules replace offending strings with <REDACTED>.
  4. Egress: The sanitized payload is securely forwarded to the external LLM provider.