Building AI Agents with Custom Actions

Install VS Code Extension

The Sema4.ai SDK provides a comprehensive framework for developing custom actions that optimize performance and ensure agents retrieve data accurately and securely, all while leveraging the full power of the Python ecosystem.

1. Open VS Code or Cursor
2. Navigate to the Extensions marketplace
3. Search for "Sema4.ai"
4. Click Install

The extension provides specialized tools for developing custom actions, including syntax highlighting, code snippets, and validation features.

Create a Custom Action

You can create a new custom action using a template:

1. Create a folder on your computer for the action
2. Open the folder in VS Code
3. Open the Command Palette (Ctrl+Shift+P or Cmd+Shift+P):
4. Type "Sema4.ai: Create Action Package" and confirm the command
5. Follow the prompts to create a basic action template

Each action package consists of four main elements contained within a specific folder structure:

• Package configuration file (package.yaml)
• Action Python code (actions.py)
• Test data inputs (.json files)
• Package icon (package.png)

Custom actions use the @action decorator to define functions that your agent can call:

@action
def calculate_metrics(data_point: float, time_period: str) -> str:
    """Calculate business metrics based on input parameters.
    
    Args:
        data_point (float): The primary value to analyze
        time_period (str): The time period for analysis (daily, weekly, monthly)
    
    Returns:
        str: Calculated metrics including trend, variance, and forecast
    """
    # Your custom logic here
    result = f"Analysis for {data_point} over {time_period} period shows positive trend"
    
    return result

Test Your Custom Action

Once you've created your action, test it with sample inputs:

1. Create a test file in the devdata folder named input_your_function_name.json (e.g., input_calculate_metrics.json)
2. Add sample input values in JSON format:

{
  "data_point": 42.5,
  "time_period": "weekly"
}

3. Open Command Pallete and type Sema4.ai: Run Action and enter
4. Review the output in the Terminal pane to ensure your action works as expected

Publish to Studio

After testing your custom action:

1. Open Command Pallete and type Sema4.ai: Publish Action Package to Sema4.ai Studio
2. Now, Sema4.ai SDK packages your action along with all dependencies to make it available for Agent's use. Once packaged, it will be pushed to Sema4.ai Studio on your machine
3. Your action package will be available in the Sema4.ai Studio Action Gallery

Implement in Your Agent

Now you can use your custom action in your agent:

1. Open Studio and navigate to your agent
2. Add the custom action from the Action Gallery
3. Update your agent's runbook to utilize the new action
4. Test your agent with the new capability