Session State Hooks

Track conversation topics in session_state by updating RunContext from a pre_hook.

Example demonstrating how to use a pre_hook to update the session_state.

session_state_hooks.py
"""
Session State Hooks
=============================

Example demonstrating how to use a pre_hook to update the session_state.
"""

from typing import List

from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.models.openai import OpenAIResponses
from agno.run import RunContext
from agno.run.agent import RunInput
from pydantic import BaseModel, Field


class ConversationTopics(BaseModel):
    topics: List[str] = Field(description="Topics present in the user messages")


# This will be our pre-hook function
def track_conversation_topics(run_context: RunContext, run_input: RunInput) -> None:
    """Simple pre-hook function to track conversation topics in the session state"""

    # Initialize the session state if it doesn't exist yet
    if run_context.session_state is None:
        run_context.session_state = {"topics": []}
    elif run_context.session_state.get("topics") is None:
        run_context.session_state["topics"] = []

    # Setup an Agent to get the topics discussed in the conversation
    # ---------------------------------------------------------------------------
    # Create Agent
    # ---------------------------------------------------------------------------

    topics_analyzer_agent = Agent(
        name="Topics Analyzer",
        model=OpenAIResponses(id="gpt-5-mini"),
        instructions=[
            "Your task is to analyze a user query and extract the topics."
            "You will be presented with a user message sent to an agent."
            "You need to extract the topics present in the user message."
            "Be concise and brief. Topics should be one or two words, and only want the one or two main topics."
            "Respond just with the list of topics, no other text or explanation."
        ],
        output_schema=ConversationTopics,
    )

    # Run the Agent to get the topics discussed in the conversation
    response = topics_analyzer_agent.run(
        input=f"Extract the topics present in the following user message: {run_input.input_content}"
    )

    # Update the session state to track the topics discussed in the conversation
    run_context.session_state["topics"].extend(response.content.topics)  # type: ignore


# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------
# Create a simple agent and equip it with our pre-hook
agent = Agent(
    name="Simple Agent",
    model=OpenAIResponses(id="gpt-5-mini"),
    pre_hooks=[track_conversation_topics],
    db=SqliteDb(db_file="test.db"),
)

# ---------------------------------------------------------------------------
# Run Agent
# ---------------------------------------------------------------------------
if __name__ == "__main__":
    agent.print_response(
        input="I want to know more about AI Agents.",
        session_id="topics_analyzer_session",
    )
    print(
        f"Current session state, after the first run: {agent.get_session_state(session_id='topics_analyzer_session')}"
    )

    agent.print_response(
        input="I also want to know more about Agno, the framework to build AI Agents.",
        session_id="topics_analyzer_session",
    )
    print(
        f"Current session state, after the second run: {agent.get_session_state(session_id='topics_analyzer_session')}"
    )

Guardrail InputCheckError and OutputCheckError become failed run outputs. They are not caught by the source's try/except around print_response() or aprint_response(). Use run() or arun() and check the returned status before displaying content or declaring success.

Add this helper after your imports:

from agno.run import RunStatus

def show_checked_response(response) -> None:
    if response.status != RunStatus.completed:
        print(f"Run rejected or failed ({response.status.value}).")
        return
    print(response.content)

A generic failed status can also indicate a provider error. Nonstream run outputs do not expose a check_trigger field. Output rejected by a post-hook can remain in the run record; this helper withholds it from the display. Streaming content may already have been emitted before the post-hook runs.

Handle a failed topic analysis

Also import CheckTrigger and InputCheckError from agno.exceptions. Immediately before extending run_context.session_state["topics"], insert this check. This version rejects the outer run if topic analysis fails, so it cannot silently claim to have updated topics.

if response.status != RunStatus.completed or not isinstance(response.content, ConversationTopics):
    raise InputCheckError(
        "The validator did not return a valid result.",
        check_trigger=CheckTrigger.INPUT_NOT_ALLOWED,
    )

After adding the helper and typed topic check, replace the final entrypoint with:

if __name__ == "__main__":
    for prompt in (
        "I want to know more about AI Agents.",
        "I also want to know more about Agno, the framework to build AI Agents.",
    ):
        response = agent.run(input=prompt, session_id="topics_analyzer_session")
        show_checked_response(response)
        if response.status == RunStatus.completed:
            print(agent.get_session_state(session_id="topics_analyzer_session"))

Run the Example

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U agno openai sqlalchemy

Export your OpenAI API key

export OPENAI_API_KEY="your_openai_api_key_here"

Apply the current checks

Add the status helper and apply the replacements described above. Keep the original imports and agent definitions that the replacement uses.

Run the example

Save the adapted code as session_state_hooks.py, then run:

python session_state_hooks.py

Full source: cookbook/02_agents/09_hooks/session_state_hooks.py