ADK Parallel Agents

234 words, 2 min read
Last updated on

ADK Parallel Agents

A ParallelAgent executes its sub-agents concurrently.

Original Notion page had a screenshot of the orchestration diagram. See migrated_from URL.

Pattern: parallel info-gathering, then sequential synthesis

A common pattern is to put a ParallelAgent (info gathering) inside a SequentialAgent (gather → synthesize).

"""
System Monitor Root Agent
 
This module defines the root agent for the system monitoring application.
It uses a parallel agent for system information gathering and a sequential
pipeline for the overall flow.
"""
 
from google.adk.agents import ParallelAgent, SequentialAgent
 
from .subagents.cpu_info_agent import cpu_info_agent
from .subagents.disk_info_agent import disk_info_agent
from .subagents.memory_info_agent import memory_info_agent
from .subagents.synthesizer_agent import system_report_synthesizer
 
# --- 1. Create Parallel Agent to gather information concurrently ---
system_info_gatherer = ParallelAgent(
    name="system_info_gatherer",
    sub_agents=[cpu_info_agent, memory_info_agent, disk_info_agent],
    # these 3 wrapped in a system_info_gatherer
)
 
# --- 2. Create Sequential Pipeline: gather info in parallel, then synthesize ---
root_agent = SequentialAgent(
    name="system_monitor_agent",
    sub_agents=[system_info_gatherer, system_report_synthesizer],
)

Keep things lightweight

Each sub-agent gets its own folder:

agent_name/
├── __pycache__/
├── __init__/
├── agent.py
└── tools.py

Sub-agents still use output_key

The synthesizer reads each parallel sub-agent’s output via state interpolation:

system_report_synthesizer = LlmAgent(
    name="SystemReportSynthesizer",
    model=GEMINI_MODEL,
    instruction="""You are a System Report Synthesizer.
 
    Your task is to create a comprehensive system health report by combining information from:
    - CPU information: {cpu_info}
    - Memory information: {memory_info}
    - Disk information: {disk_info}
    ...
    """,
)

See next