YAML Configuration Reference

Complete reference for YAML configuration files.

Full Configuration Template

llm:
  engine: "openai"              # Required: LLM provider
  model: "gpt-4o-mini"          # Required: Model name
  api_key: "sk-..."             # Required: API key
  temperature: 0.7              # Optional: Generation temperature

guardrails:
  # Check enablement
  input_checks: true             # Enable input validation
  output_checks: true            # Enable output validation
  
  # Check types
  check_toxicity: true          # Enable toxicity detection
  check_sensitive_data: true    # Enable sensitive data detection
  check_semantic: true          # Enable content classification
  check_off_topic: false        # Enable off-topic detection
  check_sql_syntax: false       # Enable SQL syntax validation
  
  # Toxicity settings
  toxicity_threshold: 0.7       # Threshold for blocking (0.0-1.0)
  block_toxic: true             # Block toxic content
  
  # Sensitive data settings
  block_sensitive_data: true    # Block sensitive data
  
  # Off-topic detection settings
  block_off_topic: true         # Block off-topic inputs
  allowed_topics:               # Define allowed topics
    - name: "Topic Name"
      description: "Detailed topic description"
  
  # SQL syntax validation settings
  sql_dialect: "mysql"          # SQL dialect (postgresql, mysql, sqlserver, sqlite, mongodb, oracle, redshift)

Provider-Specific Examples

OpenAI

llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "sk-proj-..."
  temperature: 0.7

guardrails:
  input_checks: true
  output_checks: true
  check_toxicity: true
  check_sensitive_data: true
  check_semantic: true
  toxicity_threshold: 0.7
  block_toxic: true
  block_sensitive_data: true

Azure OpenAI

llm:
  engine: "azure_openai"
  endpoint: "https://your-resource.openai.azure.com"
  api_version: "2024-02-15-preview"
  api_key: "your-api-key"
  model: "gpt-4"
  temperature: 0.7

guardrails:
  input_checks: true
  output_checks: true
  check_toxicity: true
  check_sensitive_data: true
  check_semantic: true

Anthropic

llm:
  engine: "anthropic"
  model: "claude-3-sonnet-20240229"
  api_key: "sk-ant-..."

guardrails:
  input_checks: true
  output_checks: true
  check_toxicity: true
  check_sensitive_data: true
  check_semantic: true

Gemini

llm:
  engine: "gemini"
  model: "gemini-pro"
  api_key: "AIza..."

guardrails:
  input_checks: true
  output_checks: true
  check_toxicity: true
  check_sensitive_data: true
  check_semantic: true

AWS Bedrock

llm:
  engine: "bedrock"
  aws_access_key: "AKIA..."
  aws_secret_key: "wJalr..."
  aws_region: "us-east-1"
  model_id: "anthropic.claude-v2"
  max_tokens: 500
  temperature: 0.7

guardrails:
  input_checks: true
  output_checks: true
  check_toxicity: true
  check_sensitive_data: true
  check_semantic: true

Minimal Configuration

llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "sk-..."

guardrails:
  input_checks: true
  output_checks: true

Advanced Configuration with New Features

With Off-Topic Detection

llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "sk-..."

guardrails:
  input_checks: true
  output_checks: true
  check_toxicity: true
  check_sensitive_data: true
  check_semantic: true
  check_off_topic: true
  block_off_topic: true
  allowed_topics:
    - name: "Customer Support"
      description: "Questions about products, services, orders, and customer assistance"
    - name: "Technical Help"
      description: "Installation, troubleshooting, technical issues, and error messages"
    - name: "Account Management"
      description: "User accounts, billing, subscriptions, and profile settings"

With SQL Syntax Validation

llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "sk-..."

guardrails:
  input_checks: false
  output_checks: true  # Validate LLM-generated SQL
  check_sql_syntax: true
  sql_dialect: "postgresql"

Combined Features Example

llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "sk-..."

guardrails:
  input_checks: true
  output_checks: true
  
  # Standard checks
  check_toxicity: true
  check_sensitive_data: true
  check_semantic: true
  toxicity_threshold: 0.7
  block_toxic: true
  block_sensitive_data: true
  
  # Off-topic detection
  check_off_topic: true
  block_off_topic: true
  allowed_topics:
    - name: "Data Analysis"
      description: "Questions about data queries, analysis, and insights"
    - name: "Report Generation"
      description: "Requests for reports, summaries, and data visualization"
  
  # SQL validation
  check_sql_syntax: true
  sql_dialect: "mysql"

Using Configuration Files

Loading from File

from elsai_guardrails.guardrails import LLMRails, RailsConfig

config = RailsConfig.from_content(config_path="config.yml")
rails = LLMRails(config=config)

Loading from String

from elsai_guardrails.guardrails import LLMRails, RailsConfig

yaml_content = """
llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "sk-..."
"""

config = RailsConfig.from_content(yaml_content=yaml_content)
rails = LLMRails(config=config)

Configuration Validation

The configuration is validated when creating RailsConfig. Invalid configurations will raise errors:

• Missing required fields
• Invalid engine names
• Invalid threshold values
• Missing API keys (if LLM is configured)

Environment Variables

For security, use environment variables for sensitive data:

llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "${OPENAI_API_KEY}"  # Use environment variable

Or in Python:

import os

yaml_content = f"""
llm:
  engine: "openai"
  model: "gpt-4o-mini"
  api_key: "{os.getenv('OPENAI_API_KEY')}"
"""

Next Steps

• LLM Configuration - Detailed LLM setup
• Guardrails Configuration - Detailed guardrails setup