The Adversarial Grid

A physically distributed, multi-node architecture designed for self-correction and resilience.

                    graph TD
                        subgraph "NODE 1: THE HUB (M4 Max)"
                            A[Active Mirror OS] --> B[Ollama: Think]
                            A --> C[Docker Services]
                            C --> D[SearXNG]
                            C --> E[Open-WebUI]
                        end

                        subgraph "NODE 2: ADVERSARY (M1 Max)"
                            F[Truth Scanner] --> G[Ollama: AMI]
                            F --> H[Red Team Loops]
                        end

                        subgraph "NODE 3: FIELD AGENT (Pixel 9)"
                            I[MirrorBrain Mobile] --> J[ExecuTorch Llama]
                            I --> K[Mic/Cam Sensors]
                        end

                        B <--> G
                        E <--> I
                        H -.-> A
                        
                        style A fill:#1e1e2e,stroke:#6366f1,stroke-width:2px
                        style F fill:#1e1e2e,stroke:#ef4444,stroke-width:2px
                        style I fill:#1e1e2e,stroke:#10b981,stroke-width:2px
                    

Hardware Specifications

Node 1: The Hub

Primary orchestration and heavy inference.

  • Device: MacBook Pro (M4 Max)
  • RAM: 128GB Unified
  • Role: Host of Active Mirror OS
  • Models: Llama 3.1 70B, Qwen 2.5 72B
  • Services: Docker, Vector DB, WebUI

Node 2: The Adversary

Dedicated red-teaming and truth verification.

  • Device: MacBook Pro (M1 Max)
  • RAM: 64GB Unified
  • Role: Truth Scanner & Drift Monitor
  • Models: Llama 3.2 3B (Fast checks)
  • Logic: "Trust but Verify" Loop

Node 3: Field Agent

Input collection and context gathering.

  • Device: Pixel 9 Pro
  • Chip: Google Tensor G4
  • Role: Voice/Image Input, Context
  • Models: Llama 3.2 3B (ExecuTorch)
  • OS: GrapheneOS (De-Googled)

The Mesh Network

Nodes communicate via a private, encrypted mesh network using Tailscale. This allows the "Field Agent" (Phone) to talk to the "Hub" (Desktop) securely from anywhere in the world, without exposing ports to the public internet.

  • ✓ End-to-End Encrypted (WireGuard)
  • ✓ No Open Ports
  • ✓ Direct P2P Connections
  • ✓ DNS Magic for device naming
# example_topology.json { "mesh": "tailscale0", "nodes": [ { "id": "active-mirror-hub", "ip": "100.x.y.z", "role": "master" }, { "id": "mirror-adversary", "ip": "100.a.b.c", "role": "audit" }, { "id": "mirror-mobile", "ip": "100.m.n.o", "role": "client" } ] }