Capture Reasoning Content Reasoning Tools
Capture available tool-generated reasoning from a completed run and its successful terminal streaming event.
"""
Capture Reasoning Content Reasoning Tools
=========================================
Demonstrates this reasoning cookbook example.
"""
from textwrap import dedent
from agno.agent import Agent
from agno.models.openai import OpenAIChat
from agno.tools.reasoning import ReasoningTools
# ---------------------------------------------------------------------------
# Create Example
# ---------------------------------------------------------------------------
def run_example() -> None:
"""Test function to verify reasoning_content is populated in RunOutput."""
print("\n=== Testing reasoning_content generation ===\n")
# Create an agent with ReasoningTools
agent = Agent(
model=OpenAIChat(id="gpt-5.6-luna"),
tools=[ReasoningTools(add_instructions=True)],
instructions=dedent("""\
You are an expert problem-solving assistant with strong analytical skills! Use step-by-step reasoning to solve the problem.
\
"""),
)
# Test 1: Non-streaming mode
print("Running with stream=False...")
response = agent.run(
"What is the sum of the first 10 natural numbers?", stream=False
)
# Check reasoning_content
if hasattr(response, "reasoning_content") and response.reasoning_content:
print("[OK] reasoning_content FOUND in non-streaming response")
print(f" Length: {len(response.reasoning_content)} characters")
print("\n=== reasoning_content preview (non-streaming) ===")
preview = response.reasoning_content[:1000]
if len(response.reasoning_content) > 1000:
preview += "..."
print(preview)
else:
print("[NOT FOUND] reasoning_content NOT FOUND in non-streaming response")
# Process streaming responses to find the final one
print("\n\n=== Test 2: Processing stream to find final response ===\n")
# Create another fresh agent
streaming_agent_alt = Agent(
model=OpenAIChat(id="gpt-5.6-luna"),
tools=[ReasoningTools(add_instructions=True)],
instructions=dedent("""\
You are an expert problem-solving assistant with strong analytical skills! Use step-by-step reasoning to solve the problem.
\
"""),
)
# Process streaming responses and look for the final RunOutput
final_response = None
for event in streaming_agent_alt.run(
"What is the value of 3! (factorial)?",
stream=True,
stream_events=True,
):
# The final event in the stream should be a RunOutput object
if hasattr(event, "reasoning_content"):
final_response = event
print("--- Checking reasoning_content from final stream event ---")
if (
final_response
and hasattr(final_response, "reasoning_content")
and final_response.reasoning_content
):
print("[OK] reasoning_content FOUND in final stream event")
print(f" Length: {len(final_response.reasoning_content)} characters")
print("\n=== reasoning_content preview (final stream event) ===")
preview = final_response.reasoning_content[:1000]
if len(final_response.reasoning_content) > 1000:
preview += "..."
print(preview)
else:
print("[NOT FOUND] reasoning_content NOT FOUND in final stream event")
# ---------------------------------------------------------------------------
# Run Example
# ---------------------------------------------------------------------------
if __name__ == "__main__":
run_example()Gate capture on successful completion
Reasoning content appears when the model selects the relevant tools. A reasoning event can arrive before the model later fails, so finding a reasoning_content attribute does not prove that a run completed.
Before running the preserved program, add these imports:
from agno.run.agent import RunEvent
from agno.run.base import RunStatusImmediately after the non-streaming response = agent.run(...) call, add:
if response.status != RunStatus.completed:
raise RuntimeError(f"Run ended with {response.status}")Inside the streaming loop, replace the if hasattr(event, "reasoning_content"): assignment block with:
if event.event in (RunEvent.run_error, RunEvent.run_cancelled):
raise RuntimeError(f"Stream ended with {event.event}")
if event.event == RunEvent.run_completed:
final_response = eventAfter the loop, before printing its final-result checks, add:
if final_response is None:
raise RuntimeError("Stream ended without RunCompleted")The selected object is a RunCompletedEvent, rather than a final RunOutput. Keep the existing nonempty-content check: a completed run can contain no tool-generated reasoning. These tools expose model-authored scratchpad text; their output is not a private internal trace.
Run the Example
Set up your virtual environment
uv venv --python 3.12
source .venv/bin/activateInstall dependencies
uv pip install -U agno openaiExport your OpenAI API key
export OPENAI_API_KEY="your_openai_api_key_here"Run the example
Save the code above as capture_reasoning_content_reasoning_tools.py, then run:
python capture_reasoning_content_reasoning_tools.pyFull source: cookbook/10_reasoning/tools/capture_reasoning_content_reasoning_tools.py