Dynamic Dispatch

The dispatch_task() function calls tools at template render time, enabling data-driven selection of prompts, schemas, or configuration values.

Basic Usage

In Prompts

{prompt ScenarioGenerator_prompt}
dispatch_task('handle_quiz_type')
{end_prompt}

In Schemas

- name: generate_question
  prompt: $quiz_gen.ScenarioGenerator_prompt
  schema: dispatch_task('select_output_schema')

Creating a Dispatch Tool

from agent_actions import udf_tool

@udf_tool
def handle_quiz_type(input_data: dict) -> dict:
    """Return appropriate authoring prompt based on quiz type."""
    quiz_type = input_data.get("quiz_type", "APPLICATION").upper()

    prompts = {
        "UNDERSTANDING": "Generate a conceptual question...",
        "APPLICATION": "Generate a practical question...",
    }

    return {
        "authoring_prompt": prompts.get(quiz_type, prompts["APPLICATION"]),
        "quiz_type_used": quiz_type
    }

The tool receives context (source fields, seed data, upstream outputs) and the return value replaces the dispatch_task() call.

Capturing Dispatch Results

Enable add_dispatch to preserve the dispatch tool's output in the action output:

- name: generate_question
  prompt: $quiz_gen.ScenarioGenerator_prompt
  schema: question_schema
  add_dispatch: true

When to Use Dispatch

Use dispatch when	Use Jinja2 when
Complex conditional logic	Simple if/else
Schema structure depends on input	Minor text variations
Unit-testable Python needed	Template-level logic sufficient

Basic Usage​

In Prompts​

In Schemas​

Creating a Dispatch Tool​

Capturing Dispatch Results​

When to Use Dispatch​

See Also​