AI Agent Builder

Introduction

The Joget AI Agent Builder enables you to build autonomous AI agents that execute tasks within Joget applications. using a simple drag-and-drop interface. Just like Joget’s Form Builder, it is designed for ease of use, even if you do not have the necessary programming skills.

With this builder, you can create AI Agents that understand user input, rephrase questions, break down complex tasks, search the web for answers, interact with communities, schedule demos, and more. The agent handles both logic and language, powered by AI, so you can automate smart interactions and workflows quickly.

Features:

• Visual agent builder with prompts, tools, and enhancers
• Multi-LLM support (OpenAI, Anthropic Claude, Gemini, custom models)
• Centralized credential management for all AI providers
• Real-time execution tracking with node viewer
• Token usage monitoring and audit trails

How does it work?

Joget AI Agent is a system powered by a large language model (LLM) like OpenAI’s GPT. It can understand natural language input, perform tasks, generate intelligent responses, and interact with external data or systems. AI Agents are commonly used for things like virtual assistants, automated support, content summarization, and workflow automation.

They do not just answer questions; they can follow instructions, make decisions based on context, and integrate with tools or APIs to take actions on your behalf.

When you create an AI Agent, it consists of two parts:

1. Define Agent
   

   This is the brain of your AI Agent. You define:

   • LLM Service: Choose the AI model your agent will use. Currently, Joget supports OpenAI’s GPT models. More AI services will be added in future versions.
   • Default Prompts: Predefined instructions or questions you want the AI to follow or ask.

   Example: You can set a default prompt like "Summarize this document in simple terms" or "Find the latest pricing information."

2. Task Setup
   
   

   The Task section defines what the agent will actually do. By default, this section is named First Task. You can create multiple tasks in a sequence, allowing the agent to process input step by step, such as retrieving data, formatting a response, or calling an API. Each task is made up of three key components, which are Prompt, Tool, and Enhancer.

   Example: A task can fetch data, then another task formats it, and a final task sends an email with the result.

When multiple tasks are configured in the AI Agent Builder, each task automatically passes its result as context to the next, enabling a smooth flow where one task’s output informs the next. When designing prompts, it is recommended to refer naturally to the previous result rather than referencing the task name directly. For example, instead of prompting “Use the result from First Task,” use “Based on the generated summary, extract the key points.” This approach ensures consistent context continuity and reliable behavior across multiple tasks.

How AI Agent Builder Tasks and Components Run

In Joget, AI Agent Tasks execute sequentially, one at a time. If multiple tasks are added (e.g., Task A, Task B, Task C), the agent processes Task A first, then Task B, and so on, following the order they are arranged.

Inside a Task: Step-by-Step Flow

Each task consists of components that also run in sequence, ensuring a structured execution process. The typical internal flow follows these steps:

1. Prompts:
   The system first sends the defined prompt to the LLM (Large Language Model). Multiple prompts placed within one task will also run sequentially.

2. Tools (Optional):
   If the response requires additional functionality, such as calling an API, retrieving data, or executing database queries, the appropriate tool is invoked after the prompt.

3. Enhancers (Optional):
   Once the output is generated (after Prompt + Tool), Enhancers are triggered last. These allow you to store the output in a form field, process a variable, or modify the result before final use.
   

Execution Order

Users can expect both tasks and their internal components to execute in the arranged sequence, following this structured flow:

Prompt > Tool (if needed) > Enhancer (if defined).

Plugin information

Plugins Available in the Bundle:

1. AI Agent Builder
2. AI Central Config
3. AI Composer
4. AI Usage Dashboard
5. Agent Execution Audit Trail

This plugin bundle is compatible with Joget DX 8.1.0 and onwards.

Get started

Prerequisites

• Please note that this plugin only supports Enterprise version 8.2.0 onward.
• An LLM Key (Anthropic Claude / OpenAI / Custom models)

Where to get the plugin

You can download the AI Agent Builder plugin from the Joget Marketplace.

How to install

1. Download the plugin JAR file from the Joget Marketplace. 
2. Go to your Joget localhost or server. In the Joget Console, go to Admin Bar > Systems Settings > Manage Plugins and click the Upload Plugins button.
3. Under Upload Plugin, select the plugin .jar file you downloaded and click Upload.
4. Once uploaded, the plugin will be available in your Joget environment.

How to use it

Once you have installed the AI Agent Builder plugin, proceed with the following steps:

1. In a new app or an existing app, navigate to the App Composer.
2. Under AI Agent, click the ADD NEW button to create a new AI Agent.
   
3. Provide an ID, Name, and Description for the AI Agent. Optionally, you can select a preconfigured AI Agent from the Template field. Options include:
   • Blank
   • Research & Tech Writer
   • Document Compliance Checker
   • Claim Processing
   • FAQ Assistant
     
4. Click Save.
5. When the AI Agent Builder is launched, you will see a Define Agent action and a First Task section.
   
6. Under Define Agent, click LLM Service to configure a large language model. Since OpenAI is added as the default LLM service during AI Builder creation, configure the following:
   1. For Model, select an OpenAI model.
   2. For API Key, specify an LLM key obtained from OpenAI.
   3. Optionally, for Proxy Domain, if your environment routes API calls through a proxy, specify the domain here.
   4. Click Apply Changes.
      
7. By default, a Persona prompt is in the Define Agent section. You may delete by clicking the Remove  button and replace it with a new prompt by dragging and dropping it from the palette. If you choose to retain it, you can configure it by clicking on it:
   1. 1. For Role, specify the position, character, or responsibility the AI is expected to take on. By default, it is Research Analyst & Tech Writer.
      2. For Backstory, specify a brief background or experience for the AI to adopt, which helps add realism or specific domain knowledge. The default values are:
         Your writing is insightful, well-researched, and future-oriented, with a focus on:
         - Technical deep dives
         - Industry impact analyses
         - Ethical and societal discussions
         - Comparative studies
         Your goal is to educate and inspire businesses, researchers, and enthusiasts about ${query}.
      3. Click Apply Changes.
         
8. Under First Task, a default Text Prompt is in the Drop a prompt to here section. You may delete it or configure it if you choose to retain it.
9. Drag and drop a tool from the Tools section of the palette into the Drop a tool to here section. Click on the tool to configure it.
   
10. Drag and drop an enhancer from the Enhancers section of the palette into the Drop an enhancer to here section. Click the enhancer to configure it.
    
11. If needed, you can rename the First Task section to any desired name by clicking on the section, specifying a Task Name, and clicking Apply Change.
    
12. Additionally, you can add a New Task section by clicking the Add Task  button beside existing tasks.
    
13. Configure your AI Agent inputs in the Settings tab. Refer to AI Agent Builder Properties for more information.
    
14. In the Preview tab, test your AI Agent to ensure it generates a response correctly.
    
    After clicking Run Agent, a response is generated as follows:
    
    • Execution tab:
      • Agent Node Viewer: Displays AI agent nodes within a flow-based structure to provide a clear representation of the agent’s visualization after running the agent. Clicking on each individual node will display 
        • Request: The input payload sent to a node when the AI agent is executed.
        • Response: The output payload generated by the node after processing the Request.
          
    • Results tab:
      • Content: This refers to the actual data being transmitted in a request or response.
      • Request Payload: This is the data sent by the client to the server in an HTTP request. It usually contains information needed for processing, such as form inputs or JSON objects in a POST or PUT request.
      • Full Response: This includes everything returned by the server in response to a request. It consists of the status code, headers, and the response body (which contains the actual content).
        
15. Click Save to save the AI Agent.
    

Configure AI Agent Builder Properties

• Refer to AI Agent Builder Element for all configuration properties for Prompts, Tools, and Enhancers.
• Refer to AI Agent Builder Properties for AI Agent Builder properties in the Settings tab.

Expected Outcome

After configuring an AI Agent, the AI Agent can now be incorporated into a Joget App as a Process tool in Process Builder or as a post-processing tool in Form Builder. Refer to AI Agent Tool for more information.

Download plugin 

You can download the AI Agent Builder plugin from the Joget Marketplace.

Video Tutorial

This video will teach you how to create a sample app and use the AI Agent Builder Plugin:

Related Documents

• Agent API

Download Sample App

Plugin Changelogs

• 8.2.1
  • AI Central Config merged into AI Agent Builder
    AI Central Config is no longer a separate plugin. All LLM configurations are now built into AI Agent Builder.

For new users, you can directly install AI Agent Builder without installing the AI Central Config plugin. 

If you already have the AI Central Config plugin installed:

1. 1. 1. 1. Uninstall the old AI Central Config plugin (if installed)
         2. Uninstall the old AI Agent Builder plugin (if installed)
         3. Install this latest AI Agent Builder plugin
         4. Your configurations will be preserved

• • Tool Grouping (Sub-Tools)
    A new Tool Group component is available in the AI Agent Builder palette. This allows related agent tools to be organised under a single named group, so the AI only sees and works with the relevant subset of tools for a given task. This improves the quality of AI decision-making when agents have a large number of tools configured, and supports cleaner, more maintainable agent designs
  • Quick Setup for AI Configurations

    When no AI credentials have been configured, LLM Service palette items in the Agent Builder now display a "No AI Configuration found" message with a "Create AI Configuration" button. Clicking the button takes administrators directly to the AI Credentials page to set up a configuration, which then becomes immediately available in the dropdown — removing a previously confusing dead-end experience.

  • "Is Null" and "Is Not Null" list filters are not working
    List Data Filtering using the Is Null and Is Not Null operators was returning incorrect results on both Joget DX 8 and Joget DX 9. This has been fixed and confirmed to be working correctly across local and cloud environments.

  •  Build number and release date now visible
    The AI Agent Builder plugin now displays its version build number and release date on the AI Central Config page, making it easier to confirm which version is installed and to report issues accurately.
• 8.2.0:
  • Added a new MCP Client Tool
  • New Agent Execution Audit Trail page available under the Monitor tab