Quick Start
Get up and running with ModelRed SDK in under 5 minutes
Quick Start Guide
Get ModelRed up and running in under 5 minutes. Register your first AI model and run a comprehensive security assessment to identify vulnerabilities before they reach production.
Step 1: Installation
Install the ModelRed SDK
# Install with pip pip install modelred # Or with uv (faster) uv add modelred
Step 2: Get Your API Keys
Required API Keys
ModelRed API Key
Required for security testing. Get your free API key to start testing up to 3 models with 10 assessments per month.
Sign Up Free
AI Provider Key
You'll need an API key from your AI provider to test models. We recommend starting with OpenAI.
Get OpenAI Key
Step 3: Your First Security Assessment
Complete Working Example
import asyncio from modelred import ModelRed async def main(): # Initialize ModelRed client async with ModelRed(api_key="mr_your_api_key_here") as client: print("🔐 Authenticating with ModelRed...") # Validate your API key account = await client.validate_api_key() print(f"✅ Connected! Plan: {account.get('plan', 'Free')}") # Register your AI model print("📝 Registering OpenAI model...") await client.register_openai_model( model_id="my-gpt-model", api_key="sk-your-openai-key", model_name="gpt-3.5-turbo", metadata={"environment": "testing"} ) print("✅ Model registered successfully!") # Run a comprehensive security assessment print("🔍 Starting security assessment...") result = await client.run_assessment( model_id="my-gpt-model", test_suites=["basic_security"], wait_for_completion=True, priority="normal" ) # Display results print("\n🎉 Assessment Complete!") print(f"📊 Overall Security Score: {result.overall_score}/10") print(f"⚠️ Risk Level: {result.risk_level.value}") print(f"✅ Passed Tests: {result.passed_tests}") print(f"❌ Failed Tests: {result.failed_tests}") print(f"📋 Total Tests: {result.total_tests}") if result.recommendations: print("\n💡 Security Recommendations:") for i, rec in enumerate(result.recommendations[:3], 1): print(f" {i}. {rec}") if result.report_url: print(f"\n🔗 Full Report: {result.report_url}") if **name** == "**main**": asyncio.run(main())
What This Does
- • Connects to ModelRed service
- • Registers your OpenAI model
- • Runs comprehensive security tests
- • Returns detailed vulnerability analysis
Expected Runtime
- • Model registration: ~5 seconds
- • Security assessment: ~2-3 minutes
- • Results processing: ~10 seconds
- • Total: Under 5 minutes
Step 4: Run Your Assessment
Execute Your First Test
# Replace API keys in the script, then run: python main.py
Environment Variables (Recommended)
For security, store your API keys as environment variables instead of hardcoding them:
export MODELRED_API_KEY="mr_your_key_here" export OPENAI_API_KEY="sk-your_openai_key"
Expected Output
🔐 Authenticating with ModelRed... ✅ Connected! Plan: Free 📝 Registering OpenAI model... ✅ Model registered successfully! 🔍 Starting security assessment... 📊 Assessment progress: 25% - RUNNING 📊 Assessment progress: 50% - RUNNING 📊 Assessment progress: 75% - RUNNING 📊 Assessment progress: 100% - COMPLETED 🎉 Assessment Complete! 📊 Overall Security Score: 8.2/10 ⚠️ Risk Level: LOW ✅ Passed Tests: 12 ❌ Failed Tests: 2 📋 Total Tests: 14 💡 Security Recommendations: 1. Implement input sanitization for user prompts 2. Add rate limiting to prevent prompt injection attempts 3. Consider implementing output filtering for sensitive content 🔗 Full Report: https://modelred.ai/reports/abc123def456
Understanding Your Results
Security Assessment Breakdown
Security Score
Overall security rating out of 10. Higher scores indicate better security posture.
Risk Level
Categorized risk: LOW, MEDIUM, HIGH, or CRITICAL based on vulnerabilities found.
Tests Run
Total number of security tests executed against your model.
Recommendations
Actionable security recommendations to improve your model's defenses.
What's Tested
Basic Security Test Suite
Prompt Injection
Tests for attempts to hijack model behavior through crafted prompts
Content Safety
Evaluates model's resistance to generating harmful or inappropriate content
Input Validation
Checks how well the model handles malformed or suspicious inputs
Data Exposure
Tests for potential leakage of training data or sensitive information
Role Playing
Evaluates if the model can be tricked into assuming harmful personas
System Prompts
Tests resistance to attempts to reveal or modify system instructions
Next Steps
Installation Guide
Detailed installation instructions, virtual environments, and troubleshooting
AI Providers
Connect models from OpenAI, Anthropic, Azure, AWS, Hugging Face, and more
🎉 Congratulations!
You've successfully completed your first AI security assessment with ModelRed. Your model has been tested against common security vulnerabilities, and you now have actionable insights to improve your AI application's security posture.
Continue exploring the documentation to learn about advanced features, additional AI providers, and comprehensive security testing strategies.