Custom agents

Available actions in the workflow editor

The following actions are available in the workflow editor organized by category. Each action represents a step in a workflow that an agent executes to complete a task.

Agents

Action	Usage	Example
Invoke Agent	Calls another agent to perform a task and returns its result.	Run a standard Customer Feedback Sentiment agent to process the user input before content generation. See Example: Configuring an Invoke Agent action.

Content generation

Action	Usage	Example
Generate Content	Creates new content based on instructions and context.	Generate a blog post from a product brief and target audience details.
Save Document	Saves generated content as a file or artifact.	Save a generated campaign plan as a downloadable PDF.
Compose Message	Combines inputs into a clear, user-facing message.	Turn multiple research findings into a concise executive summary.
Compose for Variant	Creates different versions of content for different audiences or conditions.	Generate separate versions of content for enterprise customers and small businesses.
Generate Image	Creates images based on prompts or content.	Generate a hero image for a blog post about AI trends.

Context parameters

Action	Usage	Example
Extract Context Params	Pulls key details from user input to use later in the workflow.	Extract company name, industry, and target audience from a user request.
Review Context Params	Checks if required information is complete and asks for missing details.	Prompt the user to provide a missing target market before continuing.
Normalize Params	Standardizes data formats so they are consistent.	Convert all dates into the same format before generating a report.
Expand Variants	Creates multiple variations from a single input.	Generate content in multiple languages or for different regions.
Retrieve Context Artifacts	Pulls in related documents or data to provide more context.	Use a product brief or document dded to the workflow to enrich generated content.

Control flow

Action	Usage	Example
For Loop	Repeats a set of steps for each item in a list.	Generate personalized emails for each contact in a list.
If/Else	Chooses different actions based on conditions.	If the user is in healthcare, generate compliant messaging; otherwise, use standard messaging.
Go To Action	Jumps to another step in the workflow.	If the step fails, go back and try again.

Data processing

Action	Usage	Example
Parse Accounts	Reads and structures account or customer data.	Extract structured account data from a file such as a CSV.
Format Enriched Accounts	Organizes enriched data into a readable format.	Turn account research into a formatted report for sales teams.
Update Document	Adds or updates content in an existing document.	Insert generated images into a marketing report.

HTTP

Action	Usage	Example
HTTP Request	Connects to an external API to fetch or send data.	Post content to an external system such as a social media platform. See Using HTTP Request in a workflow.

Human in the loop

Action	Usage	Example
Approval	Pauses the workflow until a human reviews and approves.	Require a manager to approve the generated report before finalizing.

Research

Action	Usage	Example
Deep Research	Gathers up-to-date, detailed information on a topic.	Research current trends in AI marketing tools.
Structured Research	Collects consistent, structured data across multiple items.	Gather revenue, industry, and size data for a list of companies.

Structured content

Action	Usage	Example
Get Structured Schema	Retrieves a predefined structure for content generation.	Fetch the product page schema before generating content.
Extract Variant Structure	Retrieves the structure for a specific variant.	Fetch the schema for a localized version.

Tools

Action	Usage	Example
Invoke Tool	Calls a specific Agent API tool or function within the system.	Use the Get Content tool to fetch existing page content before translating it. See Using Invoke tool in a workflow.

Variables

Action	Usage	Example
Set Variable	Stores a value to use later in the workflow.	Save the selected language to reuse in multiple steps.
Reset Variable	Clears or updates a stored value.	Reset the campaign name before starting a new workflow run.

Configuring actions

Each action must be configured before the workflow can run. This includes selecting the action type and defining its inputs, outputs, and settings.

The available configuration in the Properties pane vary by action type. Most actions include the following settings.

Input and Output variables - define how data enters the step and how results are passed to the next step.

Example: Use researchResult as input and store the output as generatedPost.
Model settings - select the model that processes the request based on speed, quality, or complexity requirements.

Example: Choose a Fast model for drafts.
Message template - defines how user inputs are converted into instructions for the AI. See best practices for writing prompts for guidance.

Templates typically include placeholders that are replaced with workflow variables (for example, {{topic}}, {{target_audience}}). They can vary between actions depending on the task.

Example: Write a LinkedIn post for a {{target_audience}} about {{topic}}.
System prompt - provides high-level instructions that guide how the AI behaves, including tone, rules, or constraints. See best practices for writing prompts for guidance.

Example: You are a senior B2B marketing strategist. Keep the tone professional and concise.
Schema and HTML template selection - select the structured output format and how it is rendered in preview.
Artifact and Storage Settings - configure how generated content is saved and identified.

Example: Save the output as text, image, or content item.

Example: Configuring an Invoke Agent action

This example demonstrates how to configure an Invoke Agent action to reuse the capabilities of a standard agent with skills within a workflow.

The Invoke Agent action allows you to call an existing standard agent as a subagent, enabling you to leverage its custom and system skills, logic, and configurations instead of recreating them. This is useful for modular workflows where existing agents can be reused across multiple processes.

Configuration:

Agent - select the standard agent you want to reuse (for example, Customer Feedback Sentiment)
Message - the input passed to the invoked agent. For example, when this is the first action after the trigger, you can pass the user’s input directly.
Model Override - use the agent’s default model or override it if needed.
Output - the response returned by the invoked agent, which can be used in subsequent workflow steps. Ensure this variable matches the input for your next step.

If you have suggestions for improving this article, let us know!