google--adk-python/contributing/new_workflow_samples/loop

ADK Workflow Loop Sample

Overview

This sample demonstrates how to create a feedback loop between different nodes in ADK Workflows.

It takes a user-provided topic and uses the generate_headline agent to write a headline. The evaluate_headline agent then grades the headline as either "tech-related" or "unrelated", providing feedback if it's unrelated. The route_headline function checks this grade. If the headline is "unrelated", the workflow loops back to the generate_headline agent, passing the feedback so it can try again. This process repeats until a "tech-related" headline is generated.

In ADK Workflows, loops allow for iterative refinement and evaluation by conditionally routing execution back to an earlier node in the sequence.

Sample Inputs

• flower
• quantum mechanics
• renewable energy

Graph

                  [ START ]
                      |
                      v
              [ process_input ]
                      |
                      v
          +->[ generate_headline ]
          |           |
          |           v
    (feedback) [ evaluate_headline ]
          |           |
          |           v
          |   [ route_headline ]
          |        /      \
          +--"unrelated"  "tech-related"
                             |
                             v
                        (Loop ends)

How To

1. Define a node (like route_headline) that yields an Event with a specific route based on a condition:

   def route_headline(node_input: Feedback):
     return Event(route=node_input.grade)
   

2. In the Workflow edges definition, create a conditional edge that connects the routing node back to a previous node in the workflow, using a routing map dict:

   (route_headline, {"unrelated": generate_headline})
   

   This creates the cycle. If the route yielded by route_headline is "unrelated", execution jumps back to generate_headline.