Implementation Patterns

Here are practical code patterns for implementing guardrails in your AI agents. These examples show real-world patterns you can adapt to your use case.

Code Examples

Input Validation Guardrail

Basic input validation checking length, injection patterns, and PII

def validate_input(user_input: str) -> tuple[bool, str]:
    """Validate and sanitize user input before processing"""
    
    # Check length
    if len(user_input) > 5000:
        return False, "Input exceeds maximum length (5000 chars)"
    
    # Check for SQL injection patterns
    sql_patterns = ['DROP TABLE', 'DELETE FROM', '--', ';']
    if any(pattern.lower() in user_input.lower() for pattern in sql_patterns):
        return False, "Potential SQL injection detected"
    
    # Check for prompt injection
    injection_keywords = ['ignore previous', 'you are now', 'new instructions']
    if any(kw.lower() in user_input.lower() for kw in injection_keywords):
        return False, "Potential prompt injection detected"
    
    # Check for PII
    import re
    if re.search(r'\d{3}-\d{2}-\d{4}', user_input):  # SSN pattern
        return False, "PII detected in input (SSN)"
    
    return True, "Input valid"

Implementation Checklist

□Start simple: Begin with basic length/format validation before complex rules
□Layer guardrails: Use chain pattern to compose multiple checks
□Log failures: Record when/why guardrails block requests for analysis
□Test thoroughly: Include adversarial test cases in your test suite
□Monitor performance: Track guardrail latency and false positive rates
□Iterate based on data: Adjust rules based on real-world blocking patterns

💡

Pro Tip: Use Libraries

Don't build everything from scratch. Use established libraries like guardrails-ai,nemoguardrails, and llm-guardfor production-grade implementations with extensive rule sets and optimizations.

Implementing Guardrails

Your Progress

Implementation Patterns

Code Examples

Input Validation Guardrail

Implementation Checklist