Configuration Fields Documentation

Basic Fields

Name and Identification

The agent needs to be configured with basic information determining its identity and operation.

Agent Name

name: Defines how the agent will be identified in conversations.

Valid examples:

• “Sales Assistant”
• “Technical Support”
• “John”
• “Maria”

Service Provider

provider: Specifies which AI service will be used.

Available providers:

• “openai” (ChatGPT)
• “anthropic” (Claude)
• “gemini” (Google)
• “deepseek” (DeepSeek)

API Key

apikey: Credential required for authentication with the chosen provider.

• Must be obtained through the selected provider’s official website
• Keep this key safe and never share it

Model Configuration

Model Selection

model: Specifies which AI model will be used. Availability depends on the selected provider.

OpenAI

Documentation: https://platform.openai.com/docs/models

• gpt-4o
• gpt-4o-mini
• gpt-3.5-turbo

Claude

Documentation: https://docs.anthropic.com/en/docs/about-claude/models

• claude-3-5-sonnet-latest
• claude-3-5-haiku-latest
• claude-3-opus-latest

Gemini

Documentation: https://ai.google.dev/models/gemini

• gemini-2.0-flash-exp
• gemini-1.5-pro
• gemini-1.5-flash

DeepSeek

Documentation: https://api-docs.deepseek.com/quick_start/pricing

• deepseek-chat

• deepseek-reasoner

Behavior Settings

Base Prompt (basePrompt)

Initial instructions to define agent behavior

Example for sales assistant:

“You are an assistant specialized in sales, focused on helping customers find the ideal products. Maintain a professional and friendly tone.”

Example for support:

“You are a technical support agent specialized in our products. Provide clear and objective answers to help customers solve their problems.”

Generation Parameters

• temperature: Controls response creativity (0-100)

  • 0-30: More conservative and precise answers

  • 30-70: Balance between creativity and precision

  • 70-100: More creative and varied answers

• maxTokens: Maximum token limit per response

  • Recommended: 1000-4000 for detailed responses

  • For short responses: 500-1000

  • Maximum limit varies by model

• diversityLevel: Controls response diversity (0-100)

  • Higher values generate more varied responses

  • Recommended: 30-70 for general use

• frequencyPenalty: Penalty for word repetition (0-100)

  • Higher values reduce repetitions

  • Recommended: 20-50 for natural communication

• presencePenalty: Penalty to maintain topic focus (0-100)

  • Higher values encourage topic changes

  • Recommended: 10-30 to maintain coherence

Interaction Settings

Messages

• signMessages: If true, adds the agent’s signature to messages

  • Useful for identifying who is sending the message

• readMessages: If true, marks messages as read when responding

  • Recommended to simulate human behavior

Configuration Examples

Sales Assistant

{
  "name": "Sales Assistant",
  "provider": "openai",
  "model": "gpt-4",
  "basePrompt": "You are a specialized sales assistant...",
  "temperature": 70,
  "maxTokens": 2000,
  "diversityLevel": 50,
  "frequencyPenalty": 30,
  "presencePenalty": 20,
  "signMessages": true,
  "readMessages": true
}

Technical Support

{
  "name": "Technical Support",
  "provider": "anthropic",
  "model": "claude-3-sonnet-20240229",
  "basePrompt": "You are a technical support agent...",
  "temperature": 30,
  "maxTokens": 3000,
  "diversityLevel": 40,
  "frequencyPenalty": 40,
  "presencePenalty": 15,
  "signMessages": true,
  "readMessages": true
}

Optimization Tips

1. Gradual Adjustment: Start with moderate values and adjust as needed

2. Test Base Prompt: Verify if instructions are clear and complete

3. Monitor Performance: Observe responses and adjust parameters for better fit

4. Backup: Keep a backup of configurations that worked well

5. Documentation: Record changes and their impacts for future reference