Compound Actions

Compound Actions empower developers to build sophisticated, multi-step automations within Agent Studio, chaining actions, data transformations, and logic into reusable AI workflows.

Think of Compound Actions as an orchestration layer that can:

• Collect input arguments from data sources
• Execute actions to complete tasks
• Conditionally run different branches of logic
• Loop over items and run steps for each item
• Run steps at the same time
• Return information to the user or processes
• Handle error gracefully
• Notify users with interactive messages and buttons

Why Use Compound Actions

• Modularity: Break down monolithic agents into testable, composable steps.
• Efficiency: Parallelize tasks or loop over lists to cut execution time.
• Resilience: Built-in error handling and early exits keep workflows robust.
• Interactivity: Notify users mid-flow with buttons for seamless handoffs.

flowchart TD
    A[Start: User Query or Trigger] --> B[Inputs<br/>e.g., Data Mapping from User/Context]
    B --> C[Core Actions<br/>e.g., HTTP Calls, Scripts]
    C --> D{Control Flow?}
    D -->|Switch: Condition Check| E[Branch 1<br/>e.g., Admin Path]
    D -->|Switch: Condition Check| F[Branch 2<br/>e.g., User Path]
    E --> G[Actions/Next Steps]
    F --> G
    G --> H[Parallel: Split Independent Tasks]
    H --> I[Task 1<br/>e.g., Fetch Data]
    H --> J[Task 2<br/>e.g., Notify User]
    I --> K[Merge Outputs]
    J --> K
    K --> L[Outputs<br/>e.g., Return Mapped Data]
    L --> M[End: Chat Response or Handoff]
    style A fill:#e1f5fe
    style M fill:#e8f5e8
    style D shape: diamond
    style H shape: diamond

The steps Key

The steps key defines a sequence of expressions executed in order, forming the backbone of multi-step Compound Actions. It groups logic cleanly, ensuring predictable flow while enabling nesting for controls like switch or parallel. Use it whenever chaining more than one expression, it's required for compound structures to maintain order and scope.

steps:
  - action:
      action_name: fetch_user_data
      output_key: user_info
      input_args:
        user_id: data.requestor_id
  - switch:  # Nested control
      cases:
        - condition: user_info.role == "admin"
          steps:
            - return:
                output_mapper:
                  access: '''granted'''

Getting Started

1. Prerequisites: Familiarity with Moveworks Data Mappers and DSL

2. Use the Low Code Editor: Get started fast with the low code editor. Switching back to code mode will generate the compound action script for you and viceversa.

   

❗️

Low Code Editor LP

The Low Code Editor is still in Limited Preview as part of the Ambient Agent LP. You can sign up for it here. Expected production release late October 2025

📘

Comments

YAML comments (#) are supported and preserved during edits, but switching between low-code and code views will strip them.