Advanced Concept

Building custom agents by directly implementing _run_async_impl provides powerful control but is more complex than using the predefined LlmAgent or standard WorkflowAgent types. We recommend understanding those foundational agent types first before tackling custom orchestration logic.

Custom agents

Custom agents provide the ultimate flexibility in ADK, allowing you to define arbitrary orchestration logic by inheriting directly from BaseAgent and implementing your own control flow. This goes beyond the predefined patterns of SequentialAgent, LoopAgent, and ParallelAgent, enabling you to build highly specific and complex agentic workflows.

Introduction: Beyond Predefined Workflows

What is a Custom Agent?

A Custom Agent is essentially any class you create that inherits from google.adk.agents.BaseAgent and implements its core execution logic within the _run_async_impl asynchronous method. You have complete control over how this method calls other agents (sub-agents), manages state, and handles events.

Why Use Them?

While the standard Workflow Agents (SequentialAgent, LoopAgent, ParallelAgent) cover common orchestration patterns, you'll need a Custom agent when your requirements include:

• Conditional Logic: Executing different sub-agents or taking different paths based on runtime conditions or the results of previous steps.
• Complex State Management: Implementing intricate logic for maintaining and updating state throughout the workflow beyond simple sequential passing.
• External Integrations: Incorporating calls to external APIs, databases, or custom Python libraries directly within the orchestration flow control.
• Dynamic Agent Selection: Choosing which sub-agent(s) to run next based on dynamic evaluation of the situation or input.
• Unique Workflow Patterns: Implementing orchestration logic that doesn't fit the standard sequential, parallel, or loop structures.

Implementing Custom Logic:

The heart of any custom agent is the _run_async_impl method. This is where you define its unique behavior.

• Signature: async def _run_async_impl(self, ctx: InvocationContext) -> AsyncGenerator[Event, None]:
• Asynchronous Generator: It must be an async def function and return an AsyncGenerator. This allows it to yield events produced by sub-agents or its own logic back to the runner.
• ctx (InvocationContext): Provides access to crucial runtime information, most importantly ctx.session.state, which is the primary way to share data between steps orchestrated by your custom agent.

Key Capabilities within _run_async_impl:

1. Calling Sub-Agents: You invoke sub-agents (which are typically stored as instance attributes like self.my_llm_agent) using their run_async method and yield their events:

   async for event in self.some_sub_agent.run_async(ctx):
       # Optionally inspect or log the event
       yield event # Pass the event up
   

2. Managing State: Read from and write to the session state dictionary (ctx.session.state) to pass data between sub-agent calls or make decisions:

   # Read data set by a previous agent
   previous_result = ctx.session.state.get("some_key")
   
   # Make a decision based on state
   if previous_result == "some_value":
       # ... call a specific sub-agent ...
   else:
       # ... call another sub-agent ...
   
   # Store a result for a later step (often done via a sub-agent's output_key)
   # ctx.session.state["my_custom_result"] = "calculated_value"
   

3. Implementing Control Flow: Use standard Python constructs (if/elif/else, for/while loops, try/except) to create sophisticated, conditional, or iterative workflows involving your sub-agents.

Managing Sub-Agents and State

Typically, a custom agent orchestrates other agents (like LlmAgent, LoopAgent, etc.).

• Initialization: You usually pass instances of these sub-agents into your custom agent's __init__ method and store them as instance attributes (e.g., self.story_generator = story_generator_instance). This makes them accessible within _run_async_impl.
• sub_agents List: When initializing the BaseAgent using super().__init__(...), you should pass a sub_agents list. This list tells the ADK framework about the agents that are part of this custom agent's immediate hierarchy. It's important for framework features like lifecycle management, introspection, and potentially future routing capabilities, even if your _run_async_impl calls the agents directly via self.xxx_agent. Include the agents that your custom logic directly invokes at the top level.
• State: As mentioned, ctx.session.state is the standard way sub-agents (especially LlmAgents using output_key) communicate results back to the orchestrator and how the orchestrator passes necessary inputs down.

Design Pattern Example: StoryFlowAgent

Let's illustrate the power of custom agents with an example pattern: a multi-stage content generation workflow with conditional logic.

Goal: Create a system that generates a story, iteratively refines it through critique and revision, performs final checks, and crucially, regenerates the story if the final tone check fails.

Why Custom? The core requirement driving the need for a custom agent here is the conditional regeneration based on the tone check. Standard workflow agents don't have built-in conditional branching based on the outcome of a sub-agent's task. We need custom Python logic (if tone == "negative": ...) within the orchestrator.

---

Part 1: Simplified custom agent Initialization

We define the StoryFlowAgent inheriting from BaseAgent. In __init__, we store the necessary sub-agents (passed in) as instance attributes and tell the BaseAgent framework about the top-level agents this custom agent will directly orchestrate.

class StoryFlowAgent(BaseAgent):
    """
    Custom agent for a story generation and refinement workflow.

    This agent orchestrates a sequence of LLM agents to generate a story,
    critique it, revise it, check grammar and tone, and potentially
    regenerate the story if the tone is negative.
    """

    # --- Field Declarations for Pydantic ---
    # Declare the agents passed during initialization as class attributes with type hints
    story_generator: LlmAgent
    critic: LlmAgent
    reviser: LlmAgent
    grammar_check: LlmAgent
    tone_check: LlmAgent

    loop_agent: LoopAgent
    sequential_agent: SequentialAgent

    # model_config allows setting Pydantic configurations if needed, e.g., arbitrary_types_allowed
    model_config = {"arbitrary_types_allowed": True}

    def __init__(
        self,
        name: str,
        story_generator: LlmAgent,
        critic: LlmAgent,
        reviser: LlmAgent,
        grammar_check: LlmAgent,
        tone_check: LlmAgent,
    ):
        """
        Initializes the StoryFlowAgent.

        Args:
            name: The name of the agent.
            story_generator: An LlmAgent to generate the initial story.
            critic: An LlmAgent to critique the story.
            reviser: An LlmAgent to revise the story based on criticism.
            grammar_check: An LlmAgent to check the grammar.
            tone_check: An LlmAgent to analyze the tone.
        """
        # Create internal agents *before* calling super().__init__
        loop_agent = LoopAgent(
            name="CriticReviserLoop", sub_agents=[critic, reviser], max_iterations=2
        )
        sequential_agent = SequentialAgent(
            name="PostProcessing", sub_agents=[grammar_check, tone_check]
        )

        # Define the sub_agents list for the framework
        sub_agents_list = [
            story_generator,
            loop_agent,
            sequential_agent,
        ]

        # Pydantic will validate and assign them based on the class annotations.
        super().__init__(
            name=name,
            story_generator=story_generator,
            critic=critic,
            reviser=reviser,
            grammar_check=grammar_check,
            tone_check=tone_check,
            loop_agent=loop_agent,
            sequential_agent=sequential_agent,
            sub_agents=sub_agents_list, # Pass the sub_agents list directly
        )

---

Part 2: Defining the Custom Execution Logic

This method orchestrates the sub-agents using standard Python async/await and control flow.

    @override
    async def _run_async_impl(
        self, ctx: InvocationContext
    ) -> AsyncGenerator[Event, None]:
        """
        Implements the custom orchestration logic for the story workflow.
        Uses the instance attributes assigned by Pydantic (e.g., self.story_generator).
        """
        logger.info(f"[{self.name}] Starting story generation workflow.")

        # 1. Initial Story Generation
        logger.info(f"[{self.name}] Running StoryGenerator...")
        async for event in self.story_generator.run_async(ctx):
            logger.info(f"[{self.name}] Event from StoryGenerator: {event.model_dump_json(indent=2, exclude_none=True)}")
            yield event

        # Check if story was generated before proceeding
        if "current_story" not in ctx.session.state or not ctx.session.state["current_story"]:
             logger.error(f"[{self.name}] Failed to generate initial story. Aborting workflow.")
             return # Stop processing if initial story failed

        logger.info(f"[{self.name}] Story state after generator: {ctx.session.state.get('current_story')}")


        # 2. Critic-Reviser Loop
        logger.info(f"[{self.name}] Running CriticReviserLoop...")
        # Use the loop_agent instance attribute assigned during init
        async for event in self.loop_agent.run_async(ctx):
            logger.info(f"[{self.name}] Event from CriticReviserLoop: {event.model_dump_json(indent=2, exclude_none=True)}")
            yield event

        logger.info(f"[{self.name}] Story state after loop: {ctx.session.state.get('current_story')}")

        # 3. Sequential Post-Processing (Grammar and Tone Check)
        logger.info(f"[{self.name}] Running PostProcessing...")
        # Use the sequential_agent instance attribute assigned during init
        async for event in self.sequential_agent.run_async(ctx):
            logger.info(f"[{self.name}] Event from PostProcessing: {event.model_dump_json(indent=2, exclude_none=True)}")
            yield event

        # 4. Tone-Based Conditional Logic
        tone_check_result = ctx.session.state.get("tone_check_result")
        logger.info(f"[{self.name}] Tone check result: {tone_check_result}")

        if tone_check_result == "negative":
            logger.info(f"[{self.name}] Tone is negative. Regenerating story...")
            async for event in self.story_generator.run_async(ctx):
                logger.info(f"[{self.name}] Event from StoryGenerator (Regen): {event.model_dump_json(indent=2, exclude_none=True)}")
                yield event
        else:
            logger.info(f"[{self.name}] Tone is not negative. Keeping current story.")
            pass

        logger.info(f"[{self.name}] Workflow finished.")

Explanation of Logic:

1. The initial story_generator runs. Its output is expected to be in ctx.session.state["current_story"].
2. The loop_agent runs, which internally calls the critic and reviser sequentially for max_iterations times. They read/write current_story and criticism from/to the state.
3. The sequential_agent runs, calling grammar_check then tone_check, reading current_story and writing grammar_suggestions and tone_check_result to the state.
4. Custom Part: The if statement checks the tone_check_result from the state. If it's "negative", the story_generator is called again, overwriting the current_story in the state. Otherwise, the flow ends.

---

Part 3: Defining the LLM Sub-Agents

These are standard LlmAgent definitions, responsible for specific tasks. Their output_key parameter is crucial for placing results into the session.state where other agents or the custom orchestrator can access them.

GEMINI_2_FLASH = "gemini-2.0-flash" # Define model constant
# --- Define the individual LLM agents ---
story_generator = LlmAgent(
    name="StoryGenerator",
    model=GEMINI_2_FLASH,
    instruction="""You are a story writer. Write a short story (around 100 words) about a cat,
based on the topic provided in session state with key 'topic'""",
    input_schema=None,
    output_key="current_story",  # Key for storing output in session state
)

critic = LlmAgent(
    name="Critic",
    model=GEMINI_2_FLASH,
    instruction="""You are a story critic. Review the story provided in
session state with key 'current_story'. Provide 1-2 sentences of constructive criticism
on how to improve it. Focus on plot or character.""",
    input_schema=None,
    output_key="criticism",  # Key for storing criticism in session state
)

reviser = LlmAgent(
    name="Reviser",
    model=GEMINI_2_FLASH,
    instruction="""You are a story reviser. Revise the story provided in
session state with key 'current_story', based on the criticism in
session state with key 'criticism'. Output only the revised story.""",
    input_schema=None,
    output_key="current_story",  # Overwrites the original story
)

grammar_check = LlmAgent(
    name="GrammarCheck",
    model=GEMINI_2_FLASH,
    instruction="""You are a grammar checker. Check the grammar of the story
provided in session state with key 'current_story'. Output only the suggested
corrections as a list, or output 'Grammar is good!' if there are no errors.""",
    input_schema=None,
    output_key="grammar_suggestions",
)

tone_check = LlmAgent(
    name="ToneCheck",
    model=GEMINI_2_FLASH,
    instruction="""You are a tone analyzer. Analyze the tone of the story
provided in session state with key 'current_story'. Output only one word: 'positive' if
the tone is generally positive, 'negative' if the tone is generally negative, or 'neutral'
otherwise.""",
    input_schema=None,
    output_key="tone_check_result", # This agent's output determines the conditional flow
)

---

Part 4: Instantiating and Running the custom agent

Finally, you instantiate your StoryFlowAgent and use the Runner as usual.

# --- Create the custom agent instance ---
story_flow_agent = StoryFlowAgent(
    name="StoryFlowAgent",
    story_generator=story_generator,
    critic=critic,
    reviser=reviser,
    grammar_check=grammar_check,
    tone_check=tone_check,
)

# --- Setup Runner and Session ---
session_service = InMemorySessionService()
initial_state = {"topic": "a brave kitten exploring a haunted house"}
session = session_service.create_session(
    app_name=APP_NAME,
    user_id=USER_ID,
    session_id=SESSION_ID,
    state=initial_state # Pass initial state here
)
logger.info(f"Initial session state: {session.state}")

runner = Runner(
    agent=story_flow_agent, # Pass the custom orchestrator agent
    app_name=APP_NAME,
    session_service=session_service
)

# --- Function to Interact with the Agent ---
def call_agent(user_input_topic: str):
    """
    Sends a new topic to the agent (overwriting the initial one if needed)
    and runs the workflow.
    """
    current_session = session_service.get_session(app_name=APP_NAME, 
                                                  user_id=USER_ID, 
                                                  session_id=SESSION_ID)
    if not current_session:
        logger.error("Session not found!")
        return

    current_session.state["topic"] = user_input_topic
    logger.info(f"Updated session state topic to: {user_input_topic}")

    content = types.Content(role='user', parts=[types.Part(text=f"Generate a story about: {user_input_topic}")])
    events = runner.run(user_id=USER_ID, session_id=SESSION_ID, new_message=content)

    final_response = "No final response captured."
    for event in events:
        if event.is_final_response() and event.content and event.content.parts:
            logger.info(f"Potential final response from [{event.author}]: {event.content.parts[0].text}")
            final_response = event.content.parts[0].text

    print("\n--- Agent Interaction Result ---")
    print("Agent Final Response: ", final_response)

    final_session = session_service.get_session(app_name=APP_NAME, 
                                                user_id=USER_ID, 
                                                session_id=SESSION_ID)
    print("Final Session State:")
    import json
    print(json.dumps(final_session.state, indent=2))
    print("-------------------------------\n")

# --- Run the Agent ---
call_agent("a lonely robot finding a friend in a junkyard")

(Note: The full runnable code, including imports and execution logic, can be found linked below.)

---

Full Code Example

Storyflow Agent

# Full runnable code for the StoryFlowAgent example
import logging
from typing import AsyncGenerator
from typing_extensions import override

from google.adk.agents import LlmAgent, BaseAgent, LoopAgent, SequentialAgent
from google.adk.agents.invocation_context import InvocationContext
from google.genai import types
from google.adk.sessions import InMemorySessionService
from google.adk.runners import Runner
from google.adk.events import Event
from pydantic import BaseModel, Field

# --- Constants ---
APP_NAME = "story_app"
USER_ID = "12345"
SESSION_ID = "123344"
GEMINI_2_FLASH = "gemini-2.0-flash"

# --- Configure Logging ---
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)


# --- Custom Orchestrator Agent ---
class StoryFlowAgent(BaseAgent):
    """
    Custom agent for a story generation and refinement workflow.

    This agent orchestrates a sequence of LLM agents to generate a story,
    critique it, revise it, check grammar and tone, and potentially
    regenerate the story if the tone is negative.
    """

    # --- Field Declarations for Pydantic ---
    # Declare the agents passed during initialization as class attributes with type hints
    story_generator: LlmAgent
    critic: LlmAgent
    reviser: LlmAgent
    grammar_check: LlmAgent
    tone_check: LlmAgent

    loop_agent: LoopAgent
    sequential_agent: SequentialAgent

    # model_config allows setting Pydantic configurations if needed, e.g., arbitrary_types_allowed
    model_config = {"arbitrary_types_allowed": True}

    def __init__(
        self,
        name: str,
        story_generator: LlmAgent,
        critic: LlmAgent,
        reviser: LlmAgent,
        grammar_check: LlmAgent,
        tone_check: LlmAgent,
    ):
        """
        Initializes the StoryFlowAgent.

        Args:
            name: The name of the agent.
            story_generator: An LlmAgent to generate the initial story.
            critic: An LlmAgent to critique the story.
            reviser: An LlmAgent to revise the story based on criticism.
            grammar_check: An LlmAgent to check the grammar.
            tone_check: An LlmAgent to analyze the tone.
        """
        # Create internal agents *before* calling super().__init__
        loop_agent = LoopAgent(
            name="CriticReviserLoop", sub_agents=[critic, reviser], max_iterations=2
        )
        sequential_agent = SequentialAgent(
            name="PostProcessing", sub_agents=[grammar_check, tone_check]
        )

        # Define the sub_agents list for the framework
        sub_agents_list = [
            story_generator,
            loop_agent,
            sequential_agent,
        ]

        # Pydantic will validate and assign them based on the class annotations.
        super().__init__(
            name=name,
            story_generator=story_generator,
            critic=critic,
            reviser=reviser,
            grammar_check=grammar_check,
            tone_check=tone_check,
            loop_agent=loop_agent,
            sequential_agent=sequential_agent,
            sub_agents=sub_agents_list, # Pass the sub_agents list directly
        )

    @override
    async def _run_async_impl(
        self, ctx: InvocationContext
    ) -> AsyncGenerator[Event, None]:
        """
        Implements the custom orchestration logic for the story workflow.
        Uses the instance attributes assigned by Pydantic (e.g., self.story_generator).
        """
        logger.info(f"[{self.name}] Starting story generation workflow.")

        # 1. Initial Story Generation
        logger.info(f"[{self.name}] Running StoryGenerator...")
        async for event in self.story_generator.run_async(ctx):
            logger.info(f"[{self.name}] Event from StoryGenerator: {event.model_dump_json(indent=2, exclude_none=True)}")
            yield event

        # Check if story was generated before proceeding
        if "current_story" not in ctx.session.state or not ctx.session.state["current_story"]:
             logger.error(f"[{self.name}] Failed to generate initial story. Aborting workflow.")
             return # Stop processing if initial story failed

        logger.info(f"[{self.name}] Story state after generator: {ctx.session.state.get('current_story')}")


        # 2. Critic-Reviser Loop
        logger.info(f"[{self.name}] Running CriticReviserLoop...")
        # Use the loop_agent instance attribute assigned during init
        async for event in self.loop_agent.run_async(ctx):
            logger.info(f"[{self.name}] Event from CriticReviserLoop: {event.model_dump_json(indent=2, exclude_none=True)}")
            yield event

        logger.info(f"[{self.name}] Story state after loop: {ctx.session.state.get('current_story')}")

        # 3. Sequential Post-Processing (Grammar and Tone Check)
        logger.info(f"[{self.name}] Running PostProcessing...")
        # Use the sequential_agent instance attribute assigned during init
        async for event in self.sequential_agent.run_async(ctx):
            logger.info(f"[{self.name}] Event from PostProcessing: {event.model_dump_json(indent=2, exclude_none=True)}")
            yield event

        # 4. Tone-Based Conditional Logic
        tone_check_result = ctx.session.state.get("tone_check_result")
        logger.info(f"[{self.name}] Tone check result: {tone_check_result}")

        if tone_check_result == "negative":
            logger.info(f"[{self.name}] Tone is negative. Regenerating story...")
            async for event in self.story_generator.run_async(ctx):
                logger.info(f"[{self.name}] Event from StoryGenerator (Regen): {event.model_dump_json(indent=2, exclude_none=True)}")
                yield event
        else:
            logger.info(f"[{self.name}] Tone is not negative. Keeping current story.")
            pass

        logger.info(f"[{self.name}] Workflow finished.")

# --- Define the individual LLM agents ---
story_generator = LlmAgent(
    name="StoryGenerator",
    model=GEMINI_2_FLASH,
    instruction="""You are a story writer. Write a short story (around 100 words) about a cat,
based on the topic provided in session state with key 'topic'""",
    input_schema=None,
    output_key="current_story",  # Key for storing output in session state
)

critic = LlmAgent(
    name="Critic",
    model=GEMINI_2_FLASH,
    instruction="""You are a story critic. Review the story provided in
session state with key 'current_story'. Provide 1-2 sentences of constructive criticism
on how to improve it. Focus on plot or character.""",
    input_schema=None,
    output_key="criticism",  # Key for storing criticism in session state
)

reviser = LlmAgent(
    name="Reviser",
    model=GEMINI_2_FLASH,
    instruction="""You are a story reviser. Revise the story provided in
session state with key 'current_story', based on the criticism in
session state with key 'criticism'. Output only the revised story.""",
    input_schema=None,
    output_key="current_story",  # Overwrites the original story
)

grammar_check = LlmAgent(
    name="GrammarCheck",
    model=GEMINI_2_FLASH,
    instruction="""You are a grammar checker. Check the grammar of the story
provided in session state with key 'current_story'. Output only the suggested
corrections as a list, or output 'Grammar is good!' if there are no errors.""",
    input_schema=None,
    output_key="grammar_suggestions",
)

tone_check = LlmAgent(
    name="ToneCheck",
    model=GEMINI_2_FLASH,
    instruction="""You are a tone analyzer. Analyze the tone of the story
provided in session state with key 'current_story'. Output only one word: 'positive' if
the tone is generally positive, 'negative' if the tone is generally negative, or 'neutral'
otherwise.""",
    input_schema=None,
    output_key="tone_check_result", # This agent's output determines the conditional flow
)

# --- Create the custom agent instance ---
story_flow_agent = StoryFlowAgent(
    name="StoryFlowAgent",
    story_generator=story_generator,
    critic=critic,
    reviser=reviser,
    grammar_check=grammar_check,
    tone_check=tone_check,
)

# --- Setup Runner and Session ---
session_service = InMemorySessionService()
initial_state = {"topic": "a brave kitten exploring a haunted house"}
session = session_service.create_session(
    app_name=APP_NAME,
    user_id=USER_ID,
    session_id=SESSION_ID,
    state=initial_state # Pass initial state here
)
logger.info(f"Initial session state: {session.state}")

runner = Runner(
    agent=story_flow_agent, # Pass the custom orchestrator agent
    app_name=APP_NAME,
    session_service=session_service
)

# --- Function to Interact with the Agent ---
def call_agent(user_input_topic: str):
    """
    Sends a new topic to the agent (overwriting the initial one if needed)
    and runs the workflow.
    """
    current_session = session_service.get_session(app_name=APP_NAME, 
                                                  user_id=USER_ID, 
                                                  session_id=SESSION_ID)
    if not current_session:
        logger.error("Session not found!")
        return

    current_session.state["topic"] = user_input_topic
    logger.info(f"Updated session state topic to: {user_input_topic}")

    content = types.Content(role='user', parts=[types.Part(text=f"Generate a story about: {user_input_topic}")])
    events = runner.run(user_id=USER_ID, session_id=SESSION_ID, new_message=content)

    final_response = "No final response captured."
    for event in events:
        if event.is_final_response() and event.content and event.content.parts:
            logger.info(f"Potential final response from [{event.author}]: {event.content.parts[0].text}")
            final_response = event.content.parts[0].text

    print("\n--- Agent Interaction Result ---")
    print("Agent Final Response: ", final_response)

    final_session = session_service.get_session(app_name=APP_NAME, 
                                                user_id=USER_ID, 
                                                session_id=SESSION_ID)
    print("Final Session State:")
    import json
    print(json.dumps(final_session.state, indent=2))
    print("-------------------------------\n")

# --- Run the Agent ---
call_agent("a lonely robot finding a friend in a junkyard")