-
Notifications
You must be signed in to change notification settings - Fork 2k
Open
Labels
agent engine[Component] This issue is related to Agent Engine deployment[Component] This issue is related to Agent Engine deploymentquestion[Component] This issue is asking a question or clarification[Component] This issue is asking a question or clarification
Description
Describe the bug
- We are getting too much of thinking from agent even for the simplest queries like "Hi"
- Is there any way to control the thoughts?
- I have tried setting thinking budget to small no. of tokens, yet it does not show any effect
To Reproduce
root_agent = Agent(
name="manager",
model=settings.GEMINI_MODEL,
description="Manager agent",
instruction=manager_instruction,
generate_content_config=GenerateContentConfig(
temperature=settings.TEMPERATURE,
http_options=HttpOptions(
# milliseconds
timeout=settings.AGENT_TIMEOUT,
),
),
tools=[
AgentTool(query_agent),
AgentTool(tax_agent),
describe_table,
explain_query,
get_schema,
sample_rows
],
output_schema=ManagerResponse,
planner=BuiltInPlanner(
thinking_config=ThinkingConfig(
include_thoughts=True, thinking_budget=settings.MANAGER_THINKING_BUDGET
)
),
)
Expected behavior
A very small thought with minimal tokens and only necessary thinking.
Screenshots


Desktop (please complete the following information):
- OS: Linux
- Python version : 3.12
- ADK version(pip show google-adk): 1.15.1
Model Information:
- Are you using LiteLLM: No
- Which model is being used : gemini-2.5-flash
Additional context
Also, can you explain why are somethings in [] , like "[briefly describe the initial instinct]". Refer the second photo
Metadata
Metadata
Assignees
Labels
agent engine[Component] This issue is related to Agent Engine deployment[Component] This issue is related to Agent Engine deploymentquestion[Component] This issue is asking a question or clarification[Component] This issue is asking a question or clarification