google--adk-python/contributing/workflow_samples/shared_sub_agent

ADK Agent Shared Sub-Agent Sample

Overview

This sample demonstrates how to reuse the same LlmAgent instance in multiple places within an agent tree.

In complex agent architectures, you may have specialized sub-agents (like a travel_agent and a shopping_agent) that both need access to the same shared capability (like a web search_agent). Instead of redefining the search_agent multiple times, you can instantiate it once and pass it as a sub-agent to multiple parents.

The ADK framework supports this naturally. When the workflow executes, events generated by the shared sub-agent will properly reflect the different execution paths (e.g., root/travel_agent/search_agent vs root/shopping_agent/search_agent).

Sample Inputs

You can test the routing and the shared sub-agent by asking travel or shopping questions:

• I want to book a flight to London from NYC.

  Routes to travel_agent -> search_agent.

• What is the price of a MacBook Air M3?

  Routes to shopping_agent -> search_agent.

• Hello, who are you?

  Handled by root_agent directly.

Graph

                  [ root_agent ]
                 /              \
                /                \
               v                  v
     [ travel_agent ]      [ shopping_agent ]
               \                  /
                \                /
                 v              v
               [ search_agent ]

How To

1. Define the Shared Agent: Instantiate the agent that will be reused.

   from google.adk.agents.llm_agent import LlmAgent
   
   search_agent = LlmAgent(
       name='search_agent',
       description='Searches the web and returns raw results.',
       instruction='...',
       disallow_transfer_to_parent=True,
       disallow_transfer_to_peers=True,
   )
   

2. Pass to Multiple Parents: Pass the exact same search_agent instance into the sub_agents list of multiple different parent agents.

   travel_agent = LlmAgent(
       name='travel_agent',
       instruction='Delegate to search_agent...',
       sub_agents=[search_agent],
   )
   
   shopping_agent = LlmAgent(
       name='shopping_agent',
       instruction='Delegate to search_agent...',
       sub_agents=[search_agent],
   )
   

3. Assemble the Root: Combine the intermediate agents under a root agent.

   root_agent = LlmAgent(
       name='root_agent',
       instruction='Route to travel_agent or shopping_agent',
       sub_agents=[travel_agent, shopping_agent],
   )