mimicxai 0.2.0

Secure foundation model management for robotics — biometrics, LLM reasoning, emotion AI, and neural mind maps.

mimicxai

Secure foundation model management for robotics — biometrics, LLM reasoning, emotion AI, and neural mind maps.

---

Overview

mimicxai is a unified Python framework for loading, running, and managing 15 production-ready AI models across three families:

Family	Models	Capabilities
Biometrix	face, iris, liveness, object_signature, voice_signature	Face/iris recognition, anti-spoofing, object fingerprinting, speaker verification
Darwin	reasoning, planning, conversation, vision, context_protection, virtual_brain	LLM reasoning via Ollama, task planning, RAG knowledge maps, sensitive data redaction
Emoticore	text, emotion_face, voice, multimodal	Emotion detection from text, faces, voice, and multimodal fusion

All models share a single Model.load() interface with built-in security (AES-256-GCM encryption, Ed25519 signing, RBAC).

Quick Start

pip install mimicxai

from mimicxai import Model

# Biometrics
face = Model.load("biometrix", task="face")
result = face.predict("photo.jpg")
match = face.match("photo_a.jpg", "photo_b.jpg")

# Darwin — LLM reasoning (auto-connects to Ollama)
darwin = Model.load("darwin", task="reasoning")
output = darwin.generate("Plan a route to the kitchen and pick up the red cup")

# Context protection — redact sensitive data
cpl = Model.load("darwin", task="context_protection")
protected = cpl.protect("SSN: 123-45-6789", provider="external")

# Virtual Brain — RAG knowledge maps
brain = Model.load("darwin", task="virtual_brain")
brain.predict(["report.pdf", "notes.txt"], action="add_files")
answer = brain.predict("What are the key findings?", action="chat")

# Emotion AI
emo = Model.load("emoticore", task="text")
mood = emo.analyze_emotion(text="I love this!")

Darwin + Ollama

Darwin LLM tasks (reasoning, planning, conversation) auto-connect to Ollama when available:

# Pull the Darwin model
ollama pull mimicxai/darwin

# Now Model.load() will use real LLM inference instead of placeholders
darwin = Model.load("darwin", task="reasoning")

Backend priority: local weights → Ollama → placeholder.

Install Extras

pip install mimicxai                    # core + registry + security
pip install mimicxai[serve]             # + FastAPI server
pip install mimicxai[vision]            # + Pillow, OpenCV
pip install mimicxai[brain]             # + scikit-learn, sentence-transformers
pip install mimicxai[train]             # + transformers, PEFT, datasets
pip install mimicxai[all]               # everything
pip install mimicxai[dev]               # + pytest, ruff, mypy, build, twine

Deployment

# Local
./deploy/deploy.sh install

# Docker Compose (API + Ollama + Nginx + Prometheus)
./deploy/deploy.sh docker up

# Kubernetes
./deploy/deploy.sh k8s apply

# systemd
sudo cp deploy/mimicxai.service /etc/systemd/system/
sudo systemctl enable --now mimicxai

Architecture

mimicxai/
├── models/
│   ├── biometrix/        # face, iris, liveness, object_signature, voice_signature
│   │   ├── face.py       # ArcFace + MTCNN (2,421 lines)
│   │   ├── iris.py       # IrisRecognition (631 lines)
│   │   ├── liveness.py   # VerifakeDetector (507 lines)
│   │   ├── object_signature.py  # ObjectBiometricNet CNN (622 lines)
│   │   └── fusion.py     # BiometricFoundationModel (706 lines)
│   ├── darwin/            # reasoning, planning, conversation, vision, CPL, virtual brain
│   │   ├── engine.py      # Ollama + transformers + GGUF (749 lines)
│   │   ├── cpl.py         # Context Protection Layer (1,085 lines)
│   │   └── virtual_brain.py  # Neural Mind Map + RAG (1,165 lines)
│   └── emoticore/         # text, face, voice, multimodal
├── security/              # AES-256-GCM, Ed25519, keyring, RBAC, audit
├── registry/              # versioning, manifests, metadata
├── serving/               # FastAPI server
├── deploy/                # Docker, K8s, systemd, Ollama, PyPI (2,500+ lines)
└── cli/                   # Typer CLI

CLI

mimicx init                              # initialize workspace
mimicx keys generate                     # Ed25519 keypair
mimicx registry add my-model ./ckpt      # register model
mimicx registry list                     # list all models
mimicx serve my-model --port 8000        # start API server

Security

Every model artifact is encrypted at rest (AES-256-GCM), signed on publish (Ed25519), and verified on load. The Context Protection Layer detects 17 sensitive data categories and enforces jurisdiction-aware policies (GDPR, CCPA, HIPAA).

License

MIT

Links

• Documentation
• GitHub
• Issues
• Changelog