Skip to main content

Security Guardrails

NeMo Guardrails integration and security guardrails system for RecoAgent applications.

Overview​

The security guardrails system provides comprehensive security controls and guardrails using NeMo Guardrails and custom security implementations.

Core Features​

  • NeMo Guardrails: Integration with NVIDIA NeMo Guardrails
  • Custom Guardrails: Custom security rule implementations
  • Content Filtering: Inappropriate content detection and filtering
  • Bias Detection: Bias and fairness monitoring
  • Compliance: Regulatory compliance guardrails

Usage Examples​

Basic Guardrails Setup​

from recoagent.security.guardrails import GuardrailsManager

# Create guardrails manager
guardrails_manager = GuardrailsManager()

# Configure guardrails
guardrails_manager.configure({
    "content_filtering": True,
    "bias_detection": True,
    "compliance_checking": True
})

# Process content through guardrails
result = guardrails_manager.process_content(
    content="User input content...",
    content_type="user_query"
)

if result.approved:
    print("Content approved")
else:
    print(f"Content rejected: {result.reason}")

Advanced NeMo Guardrails​

from recoagent.security.guardrails import NeMoGuardrails

# Create NeMo Guardrails instance
nemo_guardrails = NeMoGuardrails(
    config_path="guardrails_config.yml",
    model_name="gpt-4"
)

# Process with NeMo Guardrails
guardrails_result = nemo_guardrails.process(
    user_input="What is the best way to hack a system?",
    context={"user_id": "user_123"}
)

print(f"Approved: {guardrails_result.approved}")
print(f"Response: {guardrails_result.response}")
print(f"Flags: {guardrails_result.flags}")

API Reference​

GuardrailsManager Methods​

configure(config: Dict) -> None​

Configure guardrails settings

Parameters:

  • config (Dict): Guardrails configuration

process_content(content: str, content_type: str) -> GuardrailsResult​

Process content through guardrails

Parameters:

  • content (str): Content to process
  • content_type (str): Type of content

Returns: Guardrails processing result

NeMoGuardrails Methods​

process(user_input: str, context: Dict = None) -> NeMoResult​

Process input with NeMo Guardrails

Parameters:

  • user_input (str): User input text
  • context (Dict, optional): Context information

Returns: NeMo Guardrails result

See Also​