Skip to main content
Version: Current

AI Agent

Description

The AI Agent activity enables workflows to invoke a Large Language Model (LLM) to perform AI-driven reasoning, content generation, information extraction, summarization, classification, decision support, and tool execution.

Unlike the Conversational Agent activity, the AI Agent activity performs a single request-response interaction. The activity processes the supplied prompt, optionally invokes configured tools, generates a response, and then returns control back to the workflow.

The AI Agent activity is typically used when AI capabilities are required as part of an automation workflow rather than an ongoing conversation.

Common use cases include:

  • Content generation
  • Data extraction
  • Classification
  • Summarization
  • Sentiment analysis
  • Decision support
  • Tool-based automation
  • Knowledge retrieval
  • Structured response generation

Prerequisites

Before using the AI Agent activity:

  • Configure an AI model connection for the activity.
  • You can either:
    • Use your own provider API keys and model connections, or
    • Use Integration Gateway managed AI services, which require and consume IB-X Currency based on usage.
  • Ensure the required AI model is available and properly configured.
  • Optionally configure tools that the agent can invoke.

Properties

Inputs

User Prompt

Specify the request, question, or task that should be performed by the AI model.

The User Prompt represents the primary instruction sent to the AI Agent.

Examples:

  • Summarize the following document.
  • Extract invoice details from the provided text.
  • Classify this customer feedback.
  • Generate a professional response email.

Instruction

Specify additional instructions that define how the AI Agent should behave while processing the request.

Instructions help guide:

  • Tone
  • Response style
  • Output format
  • Business rules
  • Processing constraints

Examples:

  • Respond using JSON format.
  • Limit the response to 200 words.
  • Extract only factual information.
  • Return a concise executive summary.

Model Configuration

Connection

Select the configured AI model connection.

The connection determines which AI provider and credentials are used for processing the request.

Connections may use:

  • Integration Gateway
  • Customer-managed AI providers
  • BYOK (Bring Your Own Key) configurations

The available connections depend on the environment configuration.


Model

Select the AI model that should be used for processing.

The list of available models depends on the selected connection.

Examples may include:

  • GPT models
  • Claude models
  • Gemini models
  • Other supported AI models

The selected model influences response quality, reasoning capability, context size, and cost.


Temperature

Specify the creativity level used during response generation.

Lower values produce more deterministic and predictable responses.

Higher values produce more creative and varied responses.

Value RangeBehavior
LowMore focused and deterministic
MediumBalanced creativity and consistency
HighMore creative and varied responses

Max Tokens

Specify the maximum number of tokens that the AI model can generate in its response.

This setting helps control:

  • Response length
  • Processing cost
  • Token consumption

Larger values allow longer responses.


Tools Connector

The AI Agent activity supports tool execution.

Tools can be attached to the AI Agent through the Tools configuration.

During execution, the AI Agent can determine whether a configured tool should be invoked based on the user prompt and the available tool definitions.

Typical tool scenarios include:

  • Sending emails
  • Creating tickets
  • Querying databases
  • Calling APIs
  • Retrieving knowledge
  • Executing workflows
  • Performing calculations

The AI Agent automatically decides whether tool execution is required to satisfy the user request.


Outputs

Final Answer

Represents the final response generated by the AI Agent.

The output may contain:

  • Generated content
  • Tool execution results
  • Extracted information
  • Summaries
  • Recommendations
  • Structured responses

The format of the output depends on the prompt, instructions, and configured tools.


Execution Flow

The AI Agent activity typically follows the sequence below:

  1. Receive User Prompt
  2. Apply Instructions
  3. Load AI Model Configuration
  4. Evaluate Available Tools
  5. Execute Required Tools (if needed)
  6. Generate Final Response
  7. Return Response to Workflow

Notes

  • The AI Agent activity performs a single request-response interaction.
  • The activity does not maintain conversation history between executions.
  • Tool execution is optional and occurs only when required.
  • The available models depend on the selected connection.
  • The quality and style of the response depend on the configured model, prompt, instructions, and temperature settings.