Capabilities

What Active Mirror Does

Active Mirror is a full-stack sovereign AI system. Here's what each layer provides.

Security Layer

AMGL Guard v1.1

Pre-inference security wrapper. Blocks prompt exfiltration, role injection, and social engineering before the model ever sees the request. Fail-closed — errors block, not allow.

Request Gate

Intent classification before inference. Requests categorized and routed based on type and risk.

Drift Monitor

Continuous SHA256 verification on identity files. Automatic alerts on any mismatch.

Identity Layer

AMI Kernel v1.0

Portable identity that works across Claude, Gemini, GPT, Ollama, and any MCP-compatible model. Your AI knows who you are — everywhere.

10 Core Modules

Temporal · Emotional · Consensus · Retrieval · Vault Sync · Timeline · Mirrors · Recovery · Persona · Collaboration

Memory Layer

Semantic Vault

4,700+ documents indexed with glyph-aware embeddings. Search by meaning, not just keywords.

Truth-State Enforcement

Every claim tagged as FACT, ESTIMATE, or UNKNOWN. Vault verification required for FACT status.

Orchestration Layer

Multi-Model Routing

Use the right model for the right task. Claude for reasoning, Gemini for execution, local models for privacy.

Tri-Twin Architecture

Reflection (Claude) · Execution (Antigravity) · Anchor (Human) — three entities, one coherent operation.

Infrastructure Layer

Production Stack

  • • Mac Mini M4 (24GB) — daily production use
  • • Mac Mini M1 — adversarial testing
  • • Pixel 9 Pro (GrapheneOS) — mobile sync

State Sync

Obsidian vault → ChromaDB → iCloud. Handoffs between sessions. LaunchAgent refresh every 30 minutes.

View Full Documentation →