Skip to content

Create and Test an AI Agent

This guide demonstrates how to configure a new AI agent, including its model, tools, guardrails, and prompt. It also explains how to save the configuration and evaluate the agent using test inputs.

1. Create a New Agent

Click "Create Agent"

2. Core Configuration

The Smart Agent builder is organized into five tabs:

  1. Goal & Basics
  2. Model & Reasoning
  3. Tools & Guardrails
  4. Memory & Context
  5. Prompt & Behavior

Each tab defines a specific part of the agent.

2.1 Goal & Basics

Define the basic identity of the agent.

  • Choose a Template (e.g., Customer Support, Data Analyst)
  • Provide a clear Description
  • Add Tags for organization (e.g., "market analysis")

Choose the template as per your usecase eg. "Data Analyst"

Click the "Add a tag..." field

Click "Add"

2.2 Model & Reasoning

Select the model and configure how the agent behaves.

  • Choose a Framework (e.g., LangGraph)
  • Select a Model
  • Configure:
  • Temperature (lower = more consistent)
  • Max Steps
  • Max Tokens

Model Types

Private Models

  • Secure internal models
  • Example: claude_3_5_haiku

Org Public Models

  • Shared across the organization
  • Use when standardization is required

Model settings

2.3 Tools & Guardrails

Configure data access and safety controls.

**Tools & Guardrails**

Tools

Neo4j

  • Used to query the Knowledge Graph (Knowledge Forge)
  • Access nodes like:
  • Customer
  • Product
  • Review
  • Order

Use when: - Working with structured data - Joining multiple data sources

Other Tools

  • Web Search → External data
  • Calculator → Numeric operations
  • Web Scraper → Extract web data

MCP Server

Click "Add MCP Server"

Configure:

Transport Type
  • HTTP — request/response
  • WebSocket — real-time
  • STDIO — local communication
  • SSE — streaming
Server Details
  • Server Name
  • Remote MCP Server URL

Guardrails

Guardrails ensure safe and correct outputs.

Built-in Guardrails

  • Hallucination Detection
  • PII Detection
  • Prompt Injection Detection
  • Jailbreak Detection

Custom Guardrails

  • Add external guardrail configurations
  • Used for domain-specific rules

How to add custom guardrails

2.4 Memory & Context

Control how the agent uses memory and documents.

**Memory & Context**

Memory

  • Enable to retain session context
  • Disable for stateless execution

AI Brain (Reference Context)

  • Attach documents as context
  • Agent can use them during execution
  • All the uploaded documents in AI brain will be shown here as options

2.5 Prompt & Behavior

Define the agent’s logic using a system prompt.

  • Specify:
  • Inputs
  • Steps
  • Output format

3. Configuring Guardrails (Safety Gates)

Configure your guardrail and save

Input Guardrails

  • Filter user input before processing

Output Guardrails

  • Validate responses before display
  • Prevent:
  • PII leaks
  • Unsafe content
  • Incorrect outputs

4. Evaluate the Agent Framework

Click "Evaluate Framework"

Select frameworks

Enter test input and start evaluation

- You will see the side by side report of all frameworks , select the best performing framework as recommended.

5. Testing & Execution

Validate the agent in real time.

  • Use Test Run
  • Verify input/output behavior
  • Save configuration

Click "Langchain"

Enter your test query

View output

6. Advanced Observability: Traces

Use the Traces tab for debugging.

  • View execution steps
  • Track tool calls
  • Analyze prompt behavior