mais 2.1.2

A Jupyter-compatible plugin that detects risky ML model and dataset loads.

MAIS - ML Model Audit & Inspection System

A Python notebook plugin that watches for potentially risky model or dataset loads in Jupyter notebooks. MAIS analyzes code in real-time to detect when you're trying to load models that might require special permissions or licensing.

Detection Architecture - V1 vs V2

MAIS offers two detection architectures that can be toggled via feature flag:

🔄 V1: Legacy Baseline Detection (Default)

• Production-safe default for backward compatibility
• Uses configuration-based pattern matching
• Watches predefined function lists in config.py
• Best for: Stable production environments

🚀 V2: Provider-Based Detection (Enhanced)

• Specialized detectors for major ML/AI providers
• Comprehensive coverage including patterns V1 misses
• Provider-specific intelligence for better accuracy
• Best for: Development and comprehensive model monitoring

Provider	V1 Detection	V2 Detection
HuggingFace	✅ Basic patterns	✅ Advanced + Hub integration
OpenAI	❌ Missed patterns	✅ Full API coverage
PyTorch	✅ torch.load	✅ Extended patterns
Anthropic	❌ Not detected	✅ Claude API detection
LangChain	❌ Framework blind	✅ Full framework support
LlamaIndex	❌ Not detected	✅ Document processing

Architecture Overview

MAIS uses a flexible, strategy-based architecture with multiple specialized components:

Additional Architecture Views

View	Purpose	Link
📊 Dependencies	Component relationships & data flow	MAIS_DEPENDENCY.svg
⚡ Process Flow	End-to-end analysis workflow	MAIS_PROCESS.svg
🏗️ DDD Layers	Domain-driven design structure	MAIS_ARCHITECTURE.svg

Core Components

📥 Input Layer

Processes various types of source code inputs:

• Source Code: Direct Python code analysis
• Notebooks: Jupyter notebook cell analysis
• Requirements: Dependency file scanning
• Python Files: Static file analysis

🔍 Provider-Specific Detectors

Specialized detectors for different ML/AI providers and frameworks:

• OpenAI: Detects GPT, DALL-E, and OpenAI API usage
• HuggingFace: Identifies Transformers, Datasets, and Hub model loads
• Anthropic: Catches Claude API integrations
• LangChain: Finds LangChain components and chains
• LlamaIndex: Detects LlamaIndex document processing

⚙️ Detection Strategies

Pluggable analysis approaches that detectors can use:

• AST Strategy: Advanced parsing with variable resolution for complex code analysis
• Regex Strategy: Fast pattern matching for simple detection scenarios
• LLM-based Strategy: Future AI-powered code understanding

📊 Intermediate Output

Analysis results from provider detectors:

• Model Findings: Detected model usage with metadata
• Risk Assessment: Security and compliance evaluation
• Inventory Mapping: Model-to-provider relationship mapping

📋 JSON Schema Standardization

Converts findings into structured format:

• AI Detection JSON Schema: Standardized detection results format
• Provider Attribution: Links findings to specific ML providers
• Risk Categorization: Security and compliance classifications

📦 SBOM Generation

Creates comprehensive software bills of materials:

• manifest-cli Integration: Uses external SBOM generation tools
• SBOM Builder: Internal component for SBOM creation
• Dependency Analysis: Maps AI/ML dependencies

📤 Output Formats

Multiple standard formats for integration:

• CycloneDX JSON: Industry-standard SBOM format
• SPDX JSON: Open-source license compliance format

Installation

# Using pip
pip install mais

# Import and initialize the MAIS plugin
from mais import MAIS

# V1: Default legacy detection (production-safe)
m = MAIS(api_token="<manifest-api-token>")

# V2: Enhanced provider-based detection (recommended for dev/comprehensive monitoring)
m = MAIS(api_token="<manifest-api-token>", use_v2_detectors=True)

# Now run your notebook as normal
# MAIS will monitor for potentially risky model loads

Detection Architecture Configuration

Constructor Parameter (Per Instance)

# Use V2 provider-based detection architecture
from mais import MAIS

# Enable V2 provider-based detection (default: legacy V1)
m = MAIS(api_token="token", use_v2_detectors=True)

# Use legacy detection (default)
m = MAIS(api_token="token")  # or use_v2_detectors=False

# Explicitly use legacy detection
m = MAIS(api_token="token", use_v2_detectors=False)

Google Colab Usage

Perfect for environments where you can't set environment variables:

from google.colab import userdata
api_token = userdata.get('MANIFEST_API_KEY')

from mais import MAIS
# Use V2 for comprehensive OpenAI + HuggingFace detection
m = MAIS(api_token=api_token, use_v2_detectors=True)

Advanced Usage

MAIS supports different detection strategies and provider combinations:

from mais.application.services.ast_analyzer import ASTAnalyzer

# Use default baseline detection (backward compatible)
analyzer = ASTAnalyzer()

# Or use with custom detectors
from mais.domain.model_analysis.detectors.baseline_detector import BaselineDetector
analyzer = ASTAnalyzer(detectors=[BaselineDetector()])

# Analyze code for model usage
findings = analyzer.analyze_code(your_code)

SBOM Generation

# Generate an SBOM for your project or notebook environment.
m.create_sbom(path=".", publish=False)

SBOM Publishing

m.create_sbom(path=".", publish=True)

Environment Variables

MAIS supports configuration through environment variables:

Core Configuration

• MANIFEST_API_TOKEN - API token for MOSAIC/Manifest integration
• MAIS_MOSAIC_API_URL - Override default API URL
• MAIS_DEFAULT_VERBOSITY - Set default logging level
• MAIS_API_TIMEOUT - API request timeout in seconds

All configuration values can be overridden with MAIS_ prefix.

Detection Mode Information

from mais import MAIS

m = MAIS(api_token="token", use_v2_detectors=True)

# Check current detection mode
print(m.get_detection_mode())  # "new" or "legacy"

# Get detailed detection information
info = m.get_detection_info()
print(info["detection_mode"])      # Current mode
print(info["source"])              # "constructor parameter" or "config/environment"
print(info["feature_flag"])        # Environment variable name
print(info["current_value"])       # Boolean value of feature flag