Gemini Async Reasoning Stream

Stream available Gemini 3.7 Flash thought summaries from a separate reasoning stage.

async_reasoning_stream.py
"""
Async Reasoning Stream
======================

Demonstrates this reasoning cookbook example.
"""

import asyncio

from agno.agent import Agent
from agno.models.google import Gemini
from agno.run.agent import RunEvent  # noqa


# ---------------------------------------------------------------------------
# Create Example
# ---------------------------------------------------------------------------
def run_example() -> None:
    async def streaming_reasoning():
        """Test streaming reasoning with a Gemini model."""
        # Create an agent with reasoning enabled
        # Note: For Gemini, you MUST set thinking_budget to enable thinking mode
        agent = Agent(
            reasoning_model=Gemini(
                id="gemini-3.7-flash",
                thinking_budget=1024,  # Required to enable thinking mode
                include_thoughts=True,  # Include thought summaries in response
            ),
            instructions="Think step by step about the problem.",
        )

        prompt = "What is 25 * 37? Show your reasoning."

        await agent.aprint_response(prompt, stream=True, stream_events=True)

        # # Use manual event loop to see all events
        # async for run_output_event in agent.arun(
        #     prompt,
        #     stream=True,
        #     stream_events=True,
        # ):
        #     if run_output_event.event == RunEvent.run_started:
        #         print(f"\nEVENT: {run_output_event.event}")

        #     elif run_output_event.event == RunEvent.reasoning_started:
        #         print(f"\nEVENT: {run_output_event.event}")
        #         print("Reasoning started...\n")

        #     elif run_output_event.event == RunEvent.reasoning_content_delta:
        #         # This is the NEW streaming event for reasoning content
        #         print(run_output_event.reasoning_content, end="", flush=True)

        #     elif run_output_event.event == RunEvent.reasoning_step:
        #         print(f"\nEVENT: {run_output_event.event}")

        #     elif run_output_event.event == RunEvent.reasoning_completed:
        #         print(f"\n\nEVENT: {run_output_event.event}")

        #     elif run_output_event.event == RunEvent.run_content:
        #         if run_output_event.content:
        #             print(run_output_event.content, end="", flush=True)

        #     elif run_output_event.event == RunEvent.run_completed:
        #         print(f"\n\nEVENT: {run_output_event.event}")

    if __name__ == "__main__":
        asyncio.run(streaming_reasoning())


# ---------------------------------------------------------------------------
# Run Example
# ---------------------------------------------------------------------------
if __name__ == "__main__":
    run_example()

Gemini 3.7 thinks dynamically by default; the archived comment saying a budget is required to enable thinking is too strong. For the current levels API, replace thinking_budget=1024 with thinking_level="high", keeping include_thoughts=True. The budget field remains a compatibility setting. See Gemini thinking.

An explicit reasoning_model runs as a separate, tool-free reasoning stage before the main model response. show_full_reasoning=True displays the reasoning data Agno receives; it cannot reveal a provider's private internal trace. Some adapters use the reasoning stage's answer text when separate reasoning content is unavailable. A failed reasoning stage can still be followed by a main-model answer, so a completed run alone does not prove the reasoning stage succeeded.

Run the Example

Set up your virtual environment

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

Install dependencies

uv pip install -U agno google-genai openai

Export your API keys

export GOOGLE_API_KEY="your_google_api_key_here"
export OPENAI_API_KEY="your_openai_api_key_here"

Run the example

Save the code above as async_reasoning_stream.py, then run:

python async_reasoning_stream.py

Full source: cookbook/10_reasoning/models/gemini/async_reasoning_stream.py