Skip to main content
Feedback

Integration process design patterns for agentic workflows

Updated 

Agentic workflows in integration are compatible with conversational agents and structured agents. Refer to Structured agent mode for more information on the differences between conversational and structured agents.

You can use global variables in the Agent step and Agent Operation component. For more information, refer to Global Variables.

Requirements for structured agents

File uploads

If your agent is structured and receives a file to read, add the Data Process step to base64 encode your file. Refer to Enabling file uploads for input schema examples for agentic workflows.

Otherwise, structured agents do not require additional steps. Save your integration and select Test to test your process.

Requirements for conversational agents

If your agent is conversational, complete the following steps in your integration:

Add a Message step

Add a Message step before an Agent step to prompt the agent to respond.

  1. Add a Message step to the canvas before the Agent step.
  2. In the Message step, enter the AI agent prompt. You can copy a Conversation Starter from the agent's profile. example of the agent prompt with current data variable
  3. Optional: To pass data from a previous step:
    • In Variables, add a variable.
    • Set Type to Current Data.
  4. Select OK, then save your progress.

Add a Data Process step

The Agent step's raw output includes event notifications and metadata alongside the agent's actual message. Use a Data Process step after the Agent step to strip that out and return only the message text.

  1. Add a Data Process step to the canvas after the Agent step.
  2. Under Data Process Properties, click the plus icon to add a new property and use the following configuration:
    • Processing Step: Select Search/Replace.

    • Text to Find: [\s\S]*event: message\n.*\"content\":\s?\"(.*)\", \"timestamp\"[\s\S]*

      This searches the output of the Agent step for only the message.

    • Replace With: $1

    • Search: Select Entire document at once.

  3. Click OK to save the step.

Before adding the Data Process step, the Agent step's output includes event notifications and other information. before using the data process step, the agent step's output includes event notifications and other information

After adding the Data Process step, the output is reduced to an easy-to-read message.

after using the data process step, the agent step's output is reduced to just an easy-to-ready message

If your Agent step output still includes unexpected raw data after adding this step, refer to Agent step testing and troubleshooting.

Next steps

Test your agent. Refer to Agent step testing and troubleshooting for details.

On this Page