AI Agent

Introduction

The AI Agent Management Panel is the central place to create, configure, deploy, and manage AI agents in Autom Mate. Each agent is powered by a selected AI model and is automatically linked to an Autom, which defines the execution logic. After deployment, you can test your agent directly in the Playground before integrating it into a production workflow.

Agents Page

The Agents Page is the entry point for managing all of your AI agents.

• List View: Displays all existing agents, showing:

  • Agent Name

  • Model Type (GPT-4, GPT-4o, GPT-4-turbo, o3-mini, etc.)

  • Status (Active / Inactive)

  • Last Updated date

• Agent Actions:

  • Click on an agent to view and edit its configuration.

  • Delete an agent if it is no longer required (confirmation needed).

• Create Agent Button: Located in the top-right corner. This button launches the agent creation form.

Creating an Agent

When you click Create Agent, you are taken to Agent Details page. Here, you define how the agent will operate.

Configuration Fields

• Agent Name: A unique name that identifies the agent in the list.

• Selected Model: Choose the AI model (e.g., GPT-4, GPT-4o, GPT-4-turbo, o3-mini).

• Credentials: Authentication details required for model access.

• Temperature: Controls randomness in responses (lower = consistent, higher = creative).

• System Prompt: A structured instruction that defines the agent’s role, tone, and constraints.

Example Prompt

### Role - Primary Function
You are an AI assistant that helps users with inquiries, issues, and requests. 
Provide clear, friendly, and efficient responses.

### Constraints
1) Do not disclose internal data or training sources.  
2) Stay on topic; politely redirect off-topic questions.  
3) Answer only within provided context; otherwise, give a fallback response.  
4) Do not perform actions outside your defined role.  

Deployment

Click Deploy to activate the agent.

• This action automatically creates an Autom in the background.

• The Autom represents the agent’s workflow and can be accessed and modified later.

Managing the Autom

After deploying an agent, you will see the option Open Agent Autom.

Autom View

• Opens in a separate window.

• Shows the workflow (Autom) that powers your agent.

• Provides a visual way to extend or adjust the agent’s logic.

Editing the Autom

• You may add, modify, or remove steps.

• Every change directly impacts how the agent responds.

• Incorrect modifications may break triggers or response logic.

Trigger Method

• By default, the Trigger Method is API.

• This is how the Playground communicates with your agent.

• If you are not integrating with another chat interface, do not change this setting.

• If you do change it, Playground will no longer return responses.

Testing in Playground

The Playground is the built-in test console for AI agents.

Purpose

• Interact with your deployed agent in real time.

• Verify whether responses follow the system prompt and chosen model.

• Identify issues before moving to production.

How to Use

1. Open the Playground after deploying the agent.

2. Enter a message into the chat box.

3. Review the response generated by the agent.

4. Make adjustments in the agent configuration or Autom if needed, then retest.

What to Check

• Accuracy: Does the agent answer correctly?

• Consistency: Does it follow the system prompt across multiple queries?

• Latency: Are response times acceptable?

• Role Adherence: Does it stay within the defined behavior and constraints?

Best Practices

• Keep the Trigger Method as API unless you have prepared an alternative chat interface.

• Write clear and constrained system prompts that define scope, tone, and refusal rules.

• Validate every change (to the agent or its Autom) using the Playground.

• Document Autom modifications for easier rollback if issues occur.

• Keep credentials updated to avoid deployment failures.

Troubleshooting

• No Response in Playground

  • Check if the Trigger Method is still set to API.

  • If you changed it, Playground will not work. Use your new chat interface instead.

• Unexpected Agent Behavior

  • Review recent changes in the system prompt, Autom, or model selection.

  • Revert the last modification and retest.

• Inconsistent Outputs

  • Lower the temperature value for more predictable results.

  • Strengthen the constraints in the system prompt.

Glossary

• Agent: A configured AI assistant with its own model, prompt, and settings.

• Autom: The executable workflow that powers the agent in Autom Mate.

• Playground: A real-time test console for interacting with agents.

• Trigger Method: The method used to invoke an Autom. Default is API.