LLMSelect 1.0.0

Overview

Agent Beginner

v1.0.0 Native

Description

Allows an LLM to select an option from a list of options and execute the corresponding action.

Configuration Options

Name	Data Type	Description	Default Value
use_thread_history	bool	Whether to include previous conversation history when making the selection	false

Inputs

Name	Data Type	Description
message	list[ContentItem] or str	User message either as a string or a list of ContentItems that will be processed by the LLM to select an option

Outputs

Name	Data Type	Description
selected_option	str	The name of the option selected by the LLM, which triggers execution of the corresponding action

Examples

Basic Usage Advanced Provider-specific

# Basic customer support routing
steps:
  - name: route_inquiry
    type: LLMSelect
    config:
      llm_config:
        model: "gpt-4o"
        api_key: "sk-proj-abc123..."
        temperature: 0.3
        max_tokens: 50
        pre_prompt: "You are a customer support routing assistant. Analyze the customer inquiry and select the most appropriate department."
      use_thread_history: false
      options:
        technical_support:
          description: "Handle technical issues, bugs, and troubleshooting requests"
          action: "route_to_technical"
        billing_support:
          description: "Handle billing questions, payment issues, and account problems"
          action: "route_to_billing"
        general_inquiry:
          description: "Handle general questions and information requests"
          action: "route_to_general"
    inputs:
      message: "My payment failed and I can't access my account features"

# Advanced workflow routing with context awareness
steps:
  - name: intelligent_workflow_router
    type: LLMSelect
    config:
      llm_config:
        model: "claude-3.5-sonnet"
        api_key: "sk-ant-api03-..."
        temperature: 0.7
        max_tokens: 100
        pre_prompt: "Analyze the business process request and route to the appropriate workflow based on complexity, urgency, and resource requirements."
      use_thread_history: true
      options:
        automated_processing:
          description: "Simple, rule-based tasks that can be handled automatically without human intervention"
          action: "execute_automation"
        expert_review:
          description: "Complex cases requiring specialist knowledge and manual review"
          action: "assign_to_expert"
        collaborative_workflow:
          description: "Multi-step processes requiring coordination between multiple teams"
          action: "initiate_collaboration"
        escalation_required:
          description: "Critical issues requiring immediate management attention"
          action: "escalate_to_management"
        data_analysis_needed:
          description: "Requests requiring detailed data analysis and reporting"
          action: "route_to_analytics"
    inputs:
      message: 
        - type: "text"
          content: "We need to analyze the Q3 sales data to identify underperforming regions and create action plans"
        - type: "file" 
          path: "/data/q3_sales_report.xlsx"

# Cost-optimized content categorization with GPT-3.5
steps:
  - name: content_classifier
    type: LLMSelect
    config:
      llm_config:
        model: "gpt-3.5-turbo"
        api_key: "sk-proj-def456..."
        temperature: 0.9
        max_tokens: 30
        pre_prompt: "Classify the following content into the most appropriate category for processing."
      use_thread_history: false
      options:
        news_article:
          description: "News articles, press releases, and journalistic content"
          action: "process_news"
        marketing_content:
          description: "Promotional materials, advertisements, and marketing copy"
          action: "process_marketing"
        technical_documentation:
          description: "Technical guides, API docs, and instructional content"
          action: "process_technical"
        user_generated_content:
          description: "Reviews, comments, social media posts, and user feedback"
          action: "process_ugc"
    inputs:
      message: "Step-by-step guide to configure OAuth 2.0 authentication in your REST API application"

Error Handling

Option Selection Error

Error: LLM response did not match any of the options

Cause: The LLM's response doesn't exactly match any of the defined option names

Solution: Ensure option names are clear and descriptive. Consider using shorter, simpler names. Adjust the pre_prompt to emphasize returning exact option names only

Configuration Error

Error: LLM settings is unexpectedly None

Cause: The LLM configuration is missing required settings or the model initialization failed

Solution: Verify your llm_config contains valid model name, API key, and other required parameters. Check if the specified model is available with your API key

Token Limit Error

Error: Request exceeds maximum token limit

Cause: The input message combined with options descriptions exceeds the model's token limit

Solution: Reduce option descriptions length, shorten the pre_prompt, or use a model with higher token limits. Consider breaking complex selections into multiple steps

FAQ

How do I design effective options for LLM selection?

Create clear, distinct options that are easy for the LLM to differentiate:

• Use descriptive names: "technical_support" vs "tech" for clarity
• Provide detailed descriptions: Help the LLM understand when to select each option
• Avoid overlapping categories: Ensure options are mutually exclusive
• Keep it concise: Too many options can confuse the selection process
• Test with examples: Verify the LLM consistently selects correct options

Which LLM models work best for option selection?

Model selection depends on your specific requirements:

• GPT-4o: Best for complex decision-making with multiple criteria
• Claude-3.5-Sonnet: Excellent for nuanced categorization and context understanding
• GPT-3.5-Turbo: Cost-effective for simple binary or straightforward selections

For most routing tasks, GPT-3.5-Turbo provides sufficient accuracy at lower cost.

How should I handle selection errors and improve accuracy?

Implement strategies to improve selection reliability:

• Validate option names: Use simple, unambiguous naming conventions
• Add examples in pre_prompt: Show the LLM expected selection patterns
• Lower temperature: Use 0.1-0.3 for more consistent selections
• Fallback handling: Define default actions when no option matches
• Logging and monitoring: Track failed selections to identify patterns

When should I enable thread history vs keep it disabled?

Thread history usage depends on your workflow context:

• Enable for: Multi-turn conversations, progressive decision trees, context-aware routing
• Disable for: Independent classifications, stateless routing, high-volume processing
• Consider cost: History increases token usage and processing time
• Memory limits: Long conversations may exceed token limits

For most routing scenarios, disabling history improves performance and reduces costs.

How can I integrate LLMSelect with other workflow components?

LLMSelect works as a powerful routing mechanism in complex workflows:

• Conditional branching: Route to different processing pipelines based on content type
• Dynamic workflows: Combine with other LLM blocks for adaptive processing
• Human-in-the-loop: Route complex cases to manual review while automating simple ones
• Escalation patterns: Automatically escalate based on urgency or complexity analysis
• Load balancing: Distribute work across different processing resources

The selected option can trigger subsequent workflow steps or external system integrations.