Skip to main content
Feedback

Agent step

The Agent step allows you to integrate AI agents from the Boomi Agent Control Tower that you create with the Agent Designer directly into your integration process workflows.

To configure and use the Agent step, refer to Configuring the Agent step. For known issues and common questions, refer to Agent step troubleshooting and FAQ.

Prerequisites

  • You must have Agentstudio to configure the Agent step. Contact your Boomi representative to add this feature.
  • Your account must have a Boomi public cloud instance to use the Agent step.
  • If you install and maintain local runtimes, you must add IP addresses for communicating with Agent Tower to your allowlist. For more information, refer to Hostnames and IP addresses for the Boomi Enterprise Platform.

You can find the Agent step in the Execute section of the Steps Palette. After adding it to the canvas, you can choose from the available agents in your Control Tower to configure in the step.

Selecting an agent

The Agent step displays available agents with the following columns: Agent, Runtime, Version, and Account. Hover over any agent row to view a tooltip showing the agent's name, version, runtime, account name, and ID. Use this information to identify and select the correct agent when the same agent is deployed across multiple regions or versions.

Select the region to filter and display only agents associated with that specific region. This is useful when your account has agents across multiple regions. You can also search for a specific agent using the Search Agents search bar. To view the most recent list of agents from the Control Tower, click Refresh.

Agentstep_multi_region

  • Only the latest active deployed version of an Agent to the Runtime is displayed for selection under Agent step.
  • For the existing agents which were active before May release, the default region and Runtime is considered US and version 1.0.

Multi-region agent deployments

You can select agents deployed to multiple regions, such as ANZ, GBR, and Japan, directly within the Agent step. While you manage the actual deployments in the Agent Garden, your agents deployed outside the default US region automatically sync to the Agent Control Tower, allowing you to easily find and select them for your integration processes. For more information, refer to Agent Garden.

To help you identify integrations in your account that are ready to use with Agent step, refer to Installing recommended agents to learn more about agent templates available in Agentstudio.

Building a process around the Agent step

Before the Agent step in the workflow, you need a step that can generate a prompt for the agent, such as the Message step. You can store the Agent step output as a process property to use it elsewhere in the process.

For the step-by-step setup, refer to Configuring the Agent step.

You can route files to a structure agent directly from a process using the Agent step. This is useful when a file originates from an upstream system. For example, fetching an invoice from a file server and routing it to an agent for extraction and processing. For more information, refer to Enabling file uploads - Sending files through an agentic workflow.

Use cases

  • Customer support and service - Summarize support cases or call center interactions
  • Data analysis and insights - Analyze Boomi Community articles and comments to output a sentiment score
  • Legal and compliance assistance - Analyze prospective customer information and produce a contract draft
  • Recruitment and HR management - Input position requirements from Workday and candidate resumes from GreenHouse, and produce a score that reflects the match for next-level screening

Agent response modes

When building an agent, you can choose how an AI agent responds in the Profile screen. There are two response modes:

  • Conversational mode (default) - The AI agent responds in natural language text. This mode is ideal for conversational agents where the user may have follow-up questions after the initial prompt. The agent can use chat history and context to respond to the user and achieve its goal.

  • Structured mode - The AI agent receives JSON input from Boomi Integration and responds in a consistent, structured JSON format. This mode is ideal for workflows where the agent's response is consumed by downstream systems. The response is single-turn, meaning the agent has a single input and output with no reference to previous responses.

The mode you choose determines how you generate the configuration in the Agent step. Refer to Configuring the Agent step for the conversational and structured setup procedures.

Component Explorer

When you configure an Agent step within a process, agent-specific configurations are stored in an Agents subfolder within the process's folder in the Component Explorer.

When you create an Agent step for the first time, a root-level Agents folder is generated. This folder contains the connection details for the agent provider and cannot be edited for Boomi agents.

Each configured agent has its own subfolder within the process's Agents folder containing its specific configurations. Multiple configurations for the same agent within that process are in this subfolder.

If the same agent is used and configured in a process located in a different folder, a new Agents subfolder is created within that process's folder to hold the new configuration.

caution

Do not copy Agent connection components to another account. Connections are tied to the account in which they were created, so the copied component will include the original account information in the component URL, and the Agent step will not run.

Since you cannot edit this component, delete it and reconfigure your step to let the system recreate it with the correct information.

Limitations

  • You can configure the Agent step only with agents created in the Agent Designer

  • The Agent step accepts a single input and produces a single output, so agents with complex API tools or chatbots are not compatible

  • Context window limitations: User prompt, agent goals, tasks, and instructions combined cannot exceed 200K input tokens.

  • The Agent Designer supports XML and JSON API responses.

  • Agentstudio truncates tool responses greater than 100k tokens.

  • Regenerate configurations in the Agent step after you change the agent type in Agentstudio to reflect the updated agent definition.

  • Be cautious when updating agents in a deployed process, as these changes can cause issues or process failures.

  • You cannot create Agent connections or operations through the Component API.

  • You can continue to deploy processes to any region regardless of the Agent's deployment location in Agent Garden, as no regional restrictions are currently enforced.

  • Existing Agent Step configurations remain unchanged to ensure stability. If you have a step configured with version v1 and deploy a newer v2 through the Agent Garden, the system does not automatically upgrade your configuration. The input and output schemas only update when you explicitly select Sync with Agent Studio, and even after syncing, the version selector remains on v1 until you manually choose to change it.

  • Since multiple packaged versions share the same deployment_id, an Agent Step executes the latest deployed version (v2) even if the UI still displays v1; however, to avoid execution failures caused by schema mismatches or invalid configurations from deleted agents, you must manually sync to ensure the Agent Step reflects the current active schema.

Additional resources

On this Page