AI Governance Register

Internal AI system registry and governance tracker. Register AI systems your organization develops, deploys, or procures — with automated risk classification under the EU AI Act, impact assessments, and a governance dashboard.

Getting Started

Option 1: Hosted Version

Coming soon — a hosted instance will be available at a public URL.

Option 2: Run Locally (one command)

macOS / Linux:

./start.sh

Windows:

start.bat

This installs dependencies, starts the API and frontend, and opens your browser. No setup required beyond Python 3.11+.

Option 3: Docker

docker compose up

Then open http://localhost:8501.

Deploy Your Own

Railway (one-click cloud deploy)

1. Fork this repo
2. Connect it to Railway
3. Railway auto-detects railway.json and builds from Dockerfile.railway
4. Set environment variables in Railway dashboard:
   • ANTHROPIC_API_KEY (optional — enables AI-assisted classification)

Features

• AI System Registry — Register and manage AI systems with structured fields aligned to the EU AI Act and NIST AI RMF
• Automated Risk Classification — Rule-based classification mapped to EU AI Act Annex III categories, plus optional AI-assisted classification via Claude
• Impact Assessments — Template-generated assessments that vary by risk tier, with fill-out workflow, status tracking, and PDF/DOCX export
• Version History — Full audit trail of changes to every AI system record
• Governance Dashboard — Portfolio risk distribution, assessment completion rates, upcoming review dates, and recent activity

Risk Classification

The tool classifies AI systems into four EU AI Act risk tiers:

Tier	Examples	Obligations
Unacceptable	Social scoring, subliminal manipulation	Prohibited
High	Biometric ID, employment screening, critical infrastructure	Conformity assessment, ongoing monitoring
Limited	Chatbots, emotion recognition, deepfakes	Transparency obligations
Minimal	Content recommendation, general-purpose tools	Voluntary codes of conduct

Rule-based: Select a use case category; the system maps it to a risk tier using EU AI Act Annex III.

AI-assisted: Provide a free-text description; Claude suggests a classification with reasoning and confidence. Requires an ANTHROPIC_API_KEY.

For Developers

API Documentation

With the API running, visit http://localhost:8000/docs for interactive Swagger documentation.

Project Structure

ai-governance-register/
├── api/                        # FastAPI backend
│   ├── main.py                 # App entry, CORS, router registration
│   ├── database.py             # SQLAlchemy engine and session
│   ├── models.py               # ORM models (AISystem, ImpactAssessment, AuditLog)
│   ├── schemas.py              # Pydantic request/response schemas
│   ├── routers/                # API route modules
│   │   ├── systems.py          # AI system CRUD
│   │   ├── classification.py   # Risk classification endpoints
│   │   ├── assessments.py      # Impact assessment CRUD + export
│   │   ├── dashboard.py        # Dashboard aggregation endpoints
│   │   └── seed.py             # Sample data loader
│   └── services/               # Business logic
│       ├── risk_classifier.py  # EU AI Act rule-based classification
│       ├── llm_classifier.py   # Claude API classification
│       ├── assessment.py       # Assessment template generation
│       ├── export.py           # PDF and DOCX export
│       └── audit.py            # Version history / audit logging
├── agr_frontend/               # Streamlit frontend
│   ├── app.py                  # Main entry point
│   ├── pages/                  # Multipage app pages
│   └── utils/                  # API client, constants
├── data/sample/                # Synthetic seed data
├── tests/                      # pytest test suite
├── start.sh / start.bat        # One-command launchers
├── Dockerfile                  # API container
├── docker-compose.yml          # Multi-service compose
├── Dockerfile.railway          # Railway deployment
└── pyproject.toml              # Dependencies (PEP 621)

Running Tests

pip install -e ".[dev]"
pytest tests/ -v

Tech Stack

• Backend: Python 3.11+, FastAPI, SQLAlchemy 2.0, Pydantic v2
• Frontend: Streamlit, Plotly
• Database: SQLite (prototype), PostgreSQL-ready
• LLM: Anthropic Claude API (optional)
• Export: ReportLab (PDF), python-docx (DOCX)
• Testing: pytest
• Deployment: Docker, Railway

Environment Variables

Variable	Required	Default	Purpose
DATABASE_URL	No	sqlite:///./data/governance.db	Database connection
ANTHROPIC_API_KEY	No	—	AI-assisted classification
FRONTEND_URL	No	http://localhost:8501	CORS origin
API_URL	No	http://localhost:8000	Frontend → API URL

License

MIT