Workflow Tools
Agent drives a blog-post workflow through WorkflowTools with think and analyze steps.
"""
Workflow Tools
==============
Demonstrates this reasoning cookbook example.
"""
from textwrap import dedent
from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.models.openai import OpenAIChat
from agno.team import Team
from agno.tools.hackernews import HackerNewsTools
from agno.tools.websearch import WebSearchTools
from agno.tools.workflow import WorkflowTools
from agno.workflow.types import StepInput, StepOutput
from agno.workflow.workflow import Workflow
# ---------------------------------------------------------------------------
# Create Example
# ---------------------------------------------------------------------------
def run_example() -> None:
FEW_SHOT_EXAMPLES = dedent("""\
You can refer to the examples below as guidance for how to use each tool.
### Examples
#### Example: Blog Post Workflow
User: Please create a blog post on the topic: AI trends in 2024
Think: The user wants to process customer feedback data. I need to understand what format the data is in and what kind of summary they want. Let me start with a basic workflow run.
Run: input_data="AI trends in 2024", additional_data={"topic": "AI, AI agents, AI workflows", "style": "The blog post should be written in a style that is easy to understand and follow."}
Analyze: The workflow ran successfully and generated a basic blog post. However, the format might not be exactly what the user wants. Let me check if the results meet their expectations.
Final Answer: I've created a blog post on the topic: AI trends in 2024 through the workflow. The blog post shows...
""")
# Define agents
web_agent = Agent(
name="Web Agent",
model=OpenAIChat(id="gpt-5.6-luna"),
tools=[WebSearchTools()],
role="Search the web for the latest news and trends",
)
hackernews_agent = Agent(
name="Hackernews Agent",
model=OpenAIChat(id="gpt-5.6-luna"),
tools=[HackerNewsTools()],
role="Extract key insights and content from Hackernews posts",
)
writer_agent = Agent(
name="Writer Agent",
model=OpenAIChat(id="gpt-5.6-luna"),
instructions="Write a blog post on the topic",
)
def prepare_input_for_web_search(step_input: StepInput) -> StepOutput:
title = step_input.input
topic = step_input.additional_data.get("topic")
return StepOutput(
content=dedent(f"""\
I'm writing a blog post with the title: {title}
<topic>
{topic}
</topic>
Search the web for atleast 10 articles\
""")
)
def prepare_input_for_writer(step_input: StepInput) -> StepOutput:
title = step_input.additional_data.get("title")
topic = step_input.additional_data.get("topic")
style = step_input.additional_data.get("style")
research_team_output = step_input.previous_step_content
return StepOutput(
content=dedent(f"""\
I'm writing a blog post with the title: {title}
<required_style>
{style}
</required_style>
<topic>
{topic}
</topic>
Here is information from the web:
<research_results>
{research_team_output}
<research_results>\
""")
)
# Define research team for complex analysis
research_team = Team(
name="Research Team",
members=[hackernews_agent, web_agent],
instructions="Research tech topics from Hackernews and the web",
)
# Create and use workflow
if __name__ == "__main__":
content_creation_workflow = Workflow(
name="Blog Post Workflow",
description="Automated blog post creation from Hackernews and the web",
db=SqliteDb(
session_table="workflow_session",
db_file="tmp/workflow.db",
),
steps=[
prepare_input_for_web_search,
research_team,
prepare_input_for_writer,
writer_agent,
],
)
workflow_tools = WorkflowTools(
workflow=content_creation_workflow,
enable_think=True,
enable_analyze=True,
add_few_shot=True,
few_shot_examples=FEW_SHOT_EXAMPLES,
)
agent = Agent(
model=OpenAIChat(id="gpt-5-mini"),
tools=[workflow_tools],
markdown=True,
)
agent.print_response(
"Create a blog post with the following title: AI trends in 2024",
instructions="When you run the workflow using the `run_workflow` tool, remember to pass `additional_data` as a dictionary of key-value pairs.",
markdown=True,
stream=True,
)
# ---------------------------------------------------------------------------
# Run Example
# ---------------------------------------------------------------------------
if __name__ == "__main__":
run_example()Carry the title and optional metadata through the workflow
The archived writer preparation reads a title key absent from its own example metadata, and both preparation functions assume additional_data is present. Replace both preparation functions inside run_example() with these versions:
def prepare_input_for_web_search(step_input: StepInput) -> StepOutput:
metadata = step_input.additional_data or {}
title = step_input.get_input_as_string() or "Untitled post"
topic = metadata.get("topic") or title
return StepOutput(
content=f"Write a blog post titled: {title}\nTopic: {topic}\n"
"Search the web for at least 10 relevant articles."
)
def prepare_input_for_writer(step_input: StepInput) -> StepOutput:
metadata = step_input.additional_data or {}
title = step_input.get_input_as_string() or "Untitled post"
topic = metadata.get("topic") or title
style = metadata.get("style") or "Clear, concise prose"
research = step_input.previous_step_content or "No research returned"
return StepOutput(
content=f"Write a blog post titled: {title}\nTopic: {topic}\n"
f"Style: {style}\nResearch:\n{research}"
)The workflow's original input remains available as StepInput.input; the research result is previous_step_content. Optional metadata now refines the topic and style without needing a second copy of the title.
Move the instruction from agent.print_response(instructions=...) into the final Agent(...) constructor, replacing that constructor with:
agent = Agent(
model=OpenAIChat(id="gpt-5-mini"),
tools=[workflow_tools],
instructions=[
"Use run_workflow to create the requested blog post. "
"Its input argument is an object with input_data (the title) and "
"optional additional_data containing topic and style. "
"Inspect the returned workflow status before describing it as successful."
],
markdown=True,
)Remove the instructions=... argument from agent.print_response. That extra run keyword does not change the agent's instructions. The run_workflow tool takes a nested input object; replace the archived FEW_SHOT_EXAMPLES with:
FEW_SHOT_EXAMPLES = """To create a post titled AI trends in 2024, call run_workflow with:
{"input": {"input_data": "AI trends in 2024", "additional_data": {"topic": "AI agents and workflows", "style": "Clear introductory prose"}}}
Inspect the returned status and content before reporting the result.
"""The tools return workflow status and output as JSON to the outer agent. The outer agent finishing a response does not certify that the nested workflow succeeded.
Run the Example
Set up your virtual environment
uv venv --python 3.12
source .venv/bin/activateInstall dependencies
uv pip install -U agno ddgs fastapi openai sqlalchemyExport your OpenAI API key
export OPENAI_API_KEY="your_openai_api_key_here"Run the example
Save the code above as workflow_tools.py, then run:
python workflow_tools.pyFull source: cookbook/10_reasoning/tools/workflow_tools.py