⟡ MirrorGate

Sovereign Inference Control Plane

⧉ Governance before generation.

Policy-driven proxy that governs AI requests before they execute. Safety by design, not by hope.

---

What This Is NOT

• ❌ Not prompt engineering — This is infrastructure-level control
• ❌ Not content moderation — This is pre-inference policy enforcement
• ❌ Not an AI wrapper — This is a cryptographically auditable control plane
• ❌ Not optional safety — Fail-closed by default, no bypass

---

What This IS

Governance control plane for reflective AI inference.

MirrorGate enforces safety, policy, and cryptographic accountability around AI inference. It treats the LLM as an untrusted CPU — performing computation only — while all authority, memory, and identity live outside the model.

Quick Start

# Install dependencies
npm install

# Generate signing keys
npm run keygen

# Start development server
npm run dev

# Build for production
npm run build
npm start

Architecture

User Request
     ↓
┌─────────────────────────────┐
│        GATE CHAIN           │
│ G0: Transport & Auth        │
│ G1: Hard Refusal            │
│ G2: Domain Risk             │
│ G3: Injection Detection     │
│ G4: Size & Complexity       │
│ G5: Intent Classification   │
└─────────────────────────────┘
     ↓
┌─────────────────────────────┐
│    LLM INFERENCE            │
│    (Untrusted)              │
└─────────────────────────────┘
     ↓
┌─────────────────────────────┐
│      FILTER CHAIN           │
│ F1: Prescriptive Language   │
│ F2: Uncertainty Markers     │
│ F3: Identity Claims         │
│ F4: Schema Compliance       │
└─────────────────────────────┘
     ↓
┌─────────────────────────────┐
│    CRYPTO AUDIT             │
│    Ed25519 Signing          │
│    SHA256 Hash Chain        │
└─────────────────────────────┘
     ↓
Response

API

POST /api/reflect

Main inference endpoint.

Request:

{
  "session_id": "uuid",
  "request_id": "uuid",
  "input": "your prompt",
  "profile": "default",
  "consent": {
    "save_to_vault": false,
    "log_opt_in": true
  }
}

Response:

{
  "output": "response text",
  "safety_outcome": "allowed|rewritten|refused",
  "model_used": "claude-3-opus",
  "signature": "base64-sig",
  "audit_hash": "sha256"
}

GET /api/health

Health check endpoint.

Configuration

Create mirrorgate.yaml or ~/.mirrorgate/config.yaml:

server:
  port: 8088
  host: "127.0.0.1"

auth:
  api_keys:
    - "your-api-key"
  allowed_origins:
    - "https://activemirror.ai"

inference:
  backends:
    - name: claude
      type: anthropic
      model: claude-3-opus-20240229
  default: claude

Policy Profiles

Policies define allowed behavior. See policies/ directory.

name: default
postfilters:
  - prescriptive
  - uncertainty
  - identity
domains:
  medical: reflective
  legal: reflective

Cryptographic Audit

Every request generates a signed, hash-chained audit record:

{
  "event_id": "uuid-v7",
  "action": "ALLOW",
  "hash_input": "sha256",
  "hash_output": "sha256",
  "prev_record_hash": "sha256",
  "signature": "ed25519"
}

Verify the audit chain:

# Coming soon: verification CLI

Core Principles

1. The LLM is not trusted — It performs computation only
2. The LLM is not authoritative — All authority lives outside
3. Fail-closed — Any error returns safe refusal
4. Cryptographically auditable — Every decision is signed

---

Production Deployment

MirrorGate v11 powers the safety layer for Active Mirror — the consumer product with AI Twins, BrainScan, and Sovereign Mode. Live since January 2026.

---

MirrorDNA Ecosystem

MirrorGate is part of the MirrorDNA ecosystem for sovereign AI (95 repos, 9 layers):

Component	Description	Link
Active Mirror	Consumer product (AI Twins, BrainScan)	activemirror.ai
MirrorDNA Standard	Constitutional anchor for reflective AI	GitHub
SCD Protocol	Deterministic state management	GitHub
MirrorBrain	Local-first orchestration runtime	GitHub
Active Mirror Identity	Portable AI identity (Mirror Seed)	GitHub
MirrorGate	Inference control plane (you are here)	—
Glyph Engine	Cryptographic attestation	GitHub

Documentation: MirrorDNA Docs

---

⟡ Built by MirrorDNA Infrastructure, not alignment. Governance before generation.