# Tutorials

• Agent Left Navigation Structure
• Building an Agent with pre-trained Knowledge
• Building an Agent with Enterprise Knowledge (RAG Agent)
• Building a Team of Agents

# Agent Left Navigation Structure

The new left-side navigation panel provides an intuitive and centralized user interface for accessing all core configuration, testing, and detail management sections of the Agent.

# 1. Navigation Components

• Easy Navigation: The reorganized left-side panel ensures easy and fast access to the three primary operations required for building and managing the agent.
• Primary Sections: As highlighted in the interface, the navigation pane comprises three distinct, self-contained views:
  • Prompt
  • Benchmark
  • Agent Details
    

    

# 2. Prompt Configuration Workspace

The Prompt section serves as the main configuration hub where the agent's core conversational and operational logic is defined. This includes:

• Agent Configuration: Defining the agent’s core behavior through the System Prompt and contextual instructions (Perspective).
• Variable Definition: Specification of both Input (I/P) and Output (O/P) variables for dynamic data handling.
  • Knowledge and Tools: Integration of Knowledge Garden and external tools.
  • Testing and Debugging: A dedicated environment for real-time query testing and debugging the agent's current configuration.
  • Examples: Provision for setting up conversational examples to guide agent behavior.

# 3. Benchmark Testing

The Benchmark section is dedicated to the quantitative performance evaluation of the agent. This workspace is utilized for:

• Performance Metrics: Running standardized tests to measure the agent's accuracy, latency, and effectiveness against defined performance criteria.
• Continuous Improvement: Facilitating comparison between different agent configurations and providing data for iterative enhancements.

# 4. Agent Details (Agent Profile)

The Agent Details section, also referred to as the Agent Profile, contains all metadata and administrative settings related to the agent's identity and deployment:

• Basic Information: Configuration of the agent's Name and Description.
• Model Details: Specification of the underlying Large Language Model used by the agent.
• Input/Output Fields: Review the provided input and output variables.
• Visibility and Tags: Managing advanced settings such as access visibility (Private/Public) and applying organizational tags.

# Building an Agent with pre-trained Knowledge

# Use Case: "Marketing Idea Generator"

This example walks you through creating a Default AI Agent that generates marketing ideas for new products.

1. Step 1: Navigate to Expert Agent Studio and Start Building

   • Action: From your Purple Fabric workspace:
     • Open Expert Agent Studio.
     • Click + Build to create a new agent.

2. Step 2: Setup Agent

   • Configure basic details:
     • Solution Name: Marketing Idea Generator
     • Interaction Type: Conversation
     • Visibility: Private (or Public if intended for wider team use)

3. Step 3: Select Template

   • In the "Prompt Template" dropdown, choose Default.

4. Step 4: Define Perspective

   • In the "Perspective" field, define the agent's role, persona, goals, and guidelines.
     Example Perspective: "You are an expert marketing strategist. Your goal is to brainstorm creative and effective marketing ideas for new products. Focus on innovative campaigns, target audience engagement, and diverse channels. Be concise and provide actionable suggestions."

5. Step 5: Configure Model Settings

   • Fine-tune the agent's behavior by adjusting parameters:
     • Temperature: 0.7 (for more creative, less deterministic ideas)
     • Top-P: 0.9
     • Lookback Limit: 5 (to keep context of recent ideas)

6. Step 6: Test and Publish

   • Use the "Test & Debug" interface to run trial queries:
     Example Query: "Generate marketing ideas for a new eco-friendly smart home device." Review the generated ideas for creativity and relevance.
   • Once satisfied, review agent details and publish.

# Building an agent with Enterprise Knowledge (RAG Agent)

# Example Use Case: "Internal Policy Q&A"

This example walks you through creating a RAG Agent that answers questions based on your company's internal HR policies.

Step 1: Navigate to Expert Agent Studio and Start Building

Action: From your Purple Fabric workspace:

• Open Expert Agent Studio.
• Click + Build to create a new agent.

Step 2: Setup Agent

Configure basic details:

• Solution Name: Internal HR Policy Assistant
• Interaction Type: Conversation
• Visibility: Private (or appropriate for internal HR access)

Step 3: Select Template

In the "Prompt Template" dropdown, choose RAG.

Step 4: Add Perspective

Define the agent's role, persona, goals, and guidelines in the "Perspective" field.

• Example Perspective: "You are an HR Policy Assistant. Your role is to provide accurate and concise answers to employee questions based solely on the provided internal HR policy documents. If the answer is not in the documents, state that you cannot find the information."

Step 5: Link to Knowledge Garden

In the "Knowledge settings" section, select the EKG that contains your HR policy documents.

Step 6: Configure Retrieval Parameters

Go to "Advanced Retrieval Configurations" to fine-tune settings:

• Retrieval Strategy: Semantic Search (to find contextually relevant information)
• Chunk Size & Overlap: Adjust based on your document structure (e.g., 512 characters with 50 overlap for policy documents).
• Re-ranking: Enable for better relevance.

Step 7: Test and Publish

Thoroughly test the agent with various queries:

• Example Query 1: "What is the policy for sick leave?"
• Example Query 2: "Can I expense client entertainment?"
• Example Query 3: "What is the company's policy on remote work? (if not in documents)"

Review agent details and publish.

# Building a Team of Agents

# Example Use Case: "Customer Order Status Checker"

This example walks you through creating a ReAct Agent that uses a tool to check the status of a customer's order.

Step 1: Navigate to Expert Agent Studio and Start Building
- Action: From your Purple Fabric workspace:
- Open Expert Agent Studio.
- Click + Build to create a new agent.
- Select Team of Agents.

Step 2: Setup Agent
- Configure basic details:
- Solution Name: Customer Order Status Checker
- Interaction Type: Conversation
- Visibility: Private (or appropriate for customer service team)

Step 3: Define Perspective
- Define the role, tone, and behavior of your agent team in the "Perspective" field.
- Example Perspective: "You are a Customer Service Agent.
Your primary role is to assist customers by providing their order status. You have access to an 'Order Lookup API' tool.
Always ask for both Customer ID and Order ID if not provided.
Provide clear and concise order status updates."

Step 4: Configure Tools and Agents
- Attach the necessary tools that the ReAct agent team will use.
- Attach Tool: Select your Order Lookup API tool.
- Link any individual agents that will participate in the team's orchestration (for this simple case, no other agents are needed).

Step 5: Define ReAct Instructions
- In the "Experience" section, define the chain of thought using
the ReAct framework. This guides the agent on how to use the tool.
- Question: The user's query (e.g., "What's the status of my order? My customer ID is 123 and order ID is ABC.")
- Thought: The agent's reasoning to break down the problem.
- Example Thought: "The user wants to know their order status. I need to use the 'Order Lookup API' tool. This tool requires customer_id and order_id. Both are provided in the query."
- Action: The selection of a specific tool.
- Example Action: Order Lookup API
- Action Input: The parameters passed to the tool.
- Example Action Input: {"customer_id": "123", "order_id": "ABC"}
- Observation: The output captured from the tool. - Example Observation: {"order_status": "Shipped", "delivery_date": "2025-08-10"}
- Final Answer: The delivered result.
- Example Final Answer: "Your order ABC (Customer ID 123) has been Shipped and is expected to be delivered by August 10, 2025."
- Alternative Final Answer (if IDs missing): "I can help with that! Please provide your Customer ID and Order ID."

Step 6: Test and Publish
- Use the "Test & Debug" interface to run trial queries:
- Example Query 1: "What's the status of order XYZ for customer 456?"
- Example Query 2: "Check my order." (Verify it asks for IDs)
- Review the detailed trace steps to confirm correct tool usage and reasoning.
- Adjust Perspective, Instructions, or Tools as needed.
- Review agent details and publish.