Skip to main content

Quick Start: Agent Configuration Guide

This guide covers the essential components for configuring a FlameTree agent. Learn how to set up each component from basic identity to advanced workflow configurations, with practical examples and best practices for optimal agent performance.

Prerequisites​

  • Access to FlameTree platform
  • Basic understanding of AI agent concepts
  • YAML configuration knowledge (for advanced features)

Creating an Agent​

When you click the "Create New Agent" button and select "Custom Stateful", you'll see the configuration page with the following components:

Core Configuration Components​

Identity​

Purpose: Defines who your agent is and what organization it represents.

What to include:

  • Agent's name (e.g., "Pablo" or "Fernando")
  • Role or expertise (e.g., "Collection Expert")
  • Company representation
  • Company location or other relevant details

Example:

You are a collection expert. Your name is Pablo. You represent ABC Company. 
This company is located in New York and specializes in soft debt collection.

Best Practice: The more specific you are about the agent's identity, the more consistently it will behave in its role.

Speech Style and Language​

Purpose: Configures how your agent communicates, including language and communication style.

What to include:

  • Primary language(s) the agent should use
  • Communication style and tone
  • Speech patterns and characteristics
  • Professional level of discourse

Examples:

  • Professional: "Your language reflects the speech patterns of a professional office worker with formal vocabulary"
  • Casual: "You speak casually and friendly, like a helpful colleague"
  • Multilingual: "You speak Spanish primarily, but switch to English if the customer responds in English"

Important Note: Be specific about the style you want. Simply saying "speak like a banker" may not produce the results you expect.

Task​

Purpose: Describes the agent's primary objective and provides context for its current status.

What to include:

  • The main goal of the agent
  • Current context or situation
  • Any constraints or guidelines
  • High-level approach to the task

Example for a Collection Agent:

You are contacting a customer via WhatsApp. Your task is to collect a debt using soft collection methods (without threats or aggressive language). 
The current status is that the customer has an overdue payment of 30 days.

Workflow Configuration​

Purpose: Defines the conversation flow, states, and business logic using a state machine approach.

Key Concepts:

  • States: Different phases of the conversation
  • Transitions: Rules for moving between states
  • Tools: Available actions in each state

Complexity Levels:

  • Simple (1-2 states): Quick deployment, most basic use cases
  • Complex (3+ states): Strict control over conversation flow

Important: Workflow is the most complex configuration component. Most partners work with FlameTree support for initial setup.

→ See detailed Workflow Configuration Guide

Fast Access Knowledge Base​

Purpose: Provides quick access to information the agent needs without using external tools.

Best for: FAQ-style information that should always be available

Key Difference: This information is always available in every conversation, while the vector Knowledge Base is only searched when needed.

Example:

- What product do you offer?
- Currently, the bank offers personal loans with a credit card.
- What is a Personal Loan?
- A type of loan for personal use repaid in fixed installments.
- What is the maximum loan amount?
- USD 1,000.

Conversation Results​

Purpose: Defines what information the agent records during conversations for analysis and follow-up actions.

Format: YAML configuration specifying fields to collect

Example:

- field_name: HumanName
  kind: FormInfoField
  field_description: Human's full name
  field_type: string
  field_content: none

Where Results Are Used:

  • Session analysis and reporting
  • Operator handover information
  • Workflow decision-making
  • External system integration

Skill Groups​

Purpose: Groups of capabilities and tools available to the agent for specialized actions.

Note: Custom skills typically require working with FlameTree support for implementation.

Knowledge Base Integration​

Purpose: Provides agents with access to large amounts of domain-specific information using advanced search capabilities.

Two Types:

  • Fast Access: Always available, limited size
  • Vector Knowledge Base: Searchable database using RAG technology

→ See detailed Knowledge Base Management Guide

Configuration Best Practices​

Start Simple​

  1. Begin with basic Identity, Speech Style, and Task
  2. Use minimal workflow (1-2 states)
  3. Test thoroughly before adding complexity

Be Specific​

  • Avoid abstract descriptions like "speak like a banker"
  • Provide concrete examples of desired behavior
  • Use clear, detailed instructions

Use Examples Effectively​

  • Include dialogue examples in workflow states
  • Examples have very strong influence on agent behavior
  • Provide examples for different scenarios

Test and Iterate​

  1. Use Playground feature extensively
  2. Start with simple scenarios
  3. Gradually test edge cases
  4. Refine configuration based on results
  5. Deploy only after thorough testing

Testing Your Agent​

After configuring your agent:

  1. Use the Playground feature to test conversations
  2. Iterate on your configuration based on test results
  3. Adjust Identity, Speech Style, or Workflow as needed
  4. Test with various scenarios and edge cases
  5. Only deploy to production after thorough testing