(noob here) I am going crazy trying to implement thought summaries with Gemini 3 Flash and LiteLLM

[db-aeon]

Please see:

https://ai.google.dev/gemini-api/docs/thinking#summaries

I understand thought_signitures, but having thought_summaries are really useful in helping understand why a model may go astray. I am not seeing any documentation around this?

I am using acompletion()

Nothing I do seems to get thought in the response.

[llalitkumarrr]

Hello @db-aeon ,

To enable thinking capabilities for your models, please use the planner module from google.adk.planners (specifically BuiltInPlanner).

Example Usage:

from google.adk.planners import BuiltInPlanner

my_agent = LlmAgent(
    name="my_debugging_agent",
    model=Gemini(model="GEMINI"), # Or your desired model
    instruction="Explain your reasoning process for solving a simple math problem: 1 + 1",
    planner=BuiltInPlanner(
        thinking_config=types.ThinkingConfig(
            include_thoughts=True,
            thinking_budget=1024,
        )
    )
)

Once this is configured, the output event will include the model's thought process. You can then easily isolate and print the thoughts from the event data.