What MirrorDNA Can Do

Pre-inference blocking for prompt exfiltration, role injection, and social engineering attacks.

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

Controls external access via RESEARCH_ALLOWED flag. Air-gap mode for sensitive operations.

Continuous SHA256 checksum verification on identity files. Alerts on any mismatch.

Errors block, not allow. Unknown request types → BLOCK. Exception in validation → BLOCK.

10 modular identity components: human anchor, AI twins, context, protocols, hardware, projects, memory chain.

Same identity across Claude, Antigravity (Gemini), Ollama, and any MCP-compatible client.

SHA256 checksums on identity files. Protected fields cannot be overwritten by inference.

Automatic alerts on identity mismatch. Hard block if checksum validation fails.

Local LLM orchestration with Ollama. Identity-aware inference with vault integration.

Enforces Fact/Estimate/Unknown classification on all claims. No confident wrongness.

4,700+ documents with glyph-aware embeddings. ChromaDB vector storage.

Vault search before every response. Retrieved content takes priority over model weights.

Claude (Reflection) + Antigravity (Execution) + MirrorBrain (Local). All share same identity kernel.

Handoffs, fingerprints, LaunchAgent refresh every 30 minutes. No drift between agents.

Explicit handoffs between Claude ↔ Antigravity ↔ MirrorBrain. Summary, next actions, context preserved.

"The Dreaming" — Nightly autonomous improvement cycles. Knowledge synthesis and weight evolution.

Mac Mini M1 — Dedicated adversarial testing. Runs attack scripts against production.

Mac Mini M4 — Sovereign stack. Ollama + Open WebUI + MirrorBrain + ChromaDB.

Pixel 9 Pro (GrapheneOS) — Sovereign mobile access. Encrypted sync.

Obsidian → ChromaDB → iCloud. Triple-redundant document storage with vector embeddings.