Reasoning Reference

This reference covers the core data structures and events used across all reasoning approaches in Agno (Reasoning Models, Reasoning Tools, and Reasoning Agents).

ReasoningStep

The ReasoningStep class represents a single step in the reasoning process, whether generated by Reasoning Tools, Reasoning Agents, or native reasoning models.

Attributes

Attribute	Type	Default	Description
title	Optional[str]	None	A concise title for this reasoning step
reasoning	Optional[str]	None	The detailed thought process or reasoning for this step
action	Optional[str]	None	The action to be taken based on this reasoning
result	Optional[str]	None	The outcome or result of executing the action
next_action	NextAction	NextAction.CONTINUE	What to do next (continue, validate, final_answer, reset)
confidence	float	0.8	Confidence level for this step (0.0 to 1.0)
metadata	Optional[Dict[str, Any]]	None	Additional metadata for this step

NextAction Enum

The NextAction enum defines possible next steps in the reasoning process:

Value	Description
CONTINUE	Continue with more reasoning steps
VALIDATE	Validate the current solution before finalizing
FINAL_ANSWER	Ready to provide the final answer
RESET	Reset and restart the reasoning process (error detected)

ReasoningSteps

The ReasoningSteps class is a container for multiple reasoning steps, used as the structured output for Reasoning Agents.

Attributes

Attribute	Type	Default	Description
reasoning_steps	List[ReasoningStep]	[]	List of reasoning steps taken
metadata	Optional[Dict[str, Any]]	None	Additional metadata about the reasoning process

ReasoningTools

The ReasoningTools toolkit provides explicit tools for structured thinking.

Constructor Parameters

Parameter	Type	Default	Description
enable_think	bool	True	Enable the think() tool
enable_analyze	bool	True	Enable the analyze() tool
all	bool	False	Legacy parameter to enable both tools
instructions	Optional[str]	None	Custom instructions for using the tools
add_instructions	bool	False	Add default instructions to agent
add_few_shot	bool	False	Add few-shot examples to instructions
few_shot_examples	Optional[str]	None	Custom few-shot examples

Methods

think()

Use as a scratchpad to reason about problems step-by-step.

Parameters:

Parameter	Type	Default	Description
session_state	Dict[str, Any]	Required	Agent’s session state (automatically provided)
title	str	Required	Concise title for this thinking step
thought	str	Required	Detailed reasoning for this step
action	Optional[str]	None	What you’ll do based on this thought
confidence	float	0.8	Confidence level (0.0 to 1.0)

Returns: str - Formatted list of all reasoning steps taken so far

analyze()

Analyze results from previous actions and determine next steps.

Parameters:

Parameter	Type	Default	Description
session_state	Dict[str, Any]	Required	Agent’s session state (automatically provided)
title	str	Required	Concise title for this analysis
result	str	Required	Outcome of the previous action
analysis	str	Required	Your evaluation of the results
next_action	str	"continue"	What to do next: “continue”, “validate”, or “final_answer”
confidence	float	0.8	Confidence level (0.0 to 1.0)

Returns: str - Formatted list of all reasoning steps taken so far

Reasoning Events

Events emitted during reasoning processes when using Reasoning Agents or Reasoning Models.

Event Types

Event Type	Description
ReasoningStarted	Indicates the start of the reasoning process
ReasoningStep	Contains a single reasoning step
ReasoningCompleted	Signals completion of the reasoning process

ReasoningStartedEvent

Emitted when reasoning begins.

Attributes:

Attribute	Type	Default	Description
event	str	"ReasoningStarted"	Event type
run_id	Optional[str]	None	ID of the current run
agent_id	Optional[str]	None	ID of the reasoning agent
created_at	int	Current timestamp	Unix timestamp of event creation

ReasoningStepEvent

Emitted for each reasoning step during the process.

Attributes:

Attribute	Type	Default	Description
event	str	"ReasoningStep"	Event type
content	Optional[Any]	None	Content of the reasoning step
content_type	str	"str"	Type of the content
reasoning_content	str	""	Detailed reasoning content for this step
run_id	Optional[str]	None	ID of the current run
agent_id	Optional[str]	None	ID of the reasoning agent
step_number	Optional[int]	None	The sequential number of this step
created_at	int	Current timestamp	Unix timestamp of event creation

ReasoningCompletedEvent

Emitted when reasoning finishes.

Attributes:

Attribute	Type	Default	Description
event	str	"ReasoningCompleted"	Event type
content	Optional[Any]	None	Final reasoning content
content_type	str	"str"	Type of the content
reasoning_steps	Optional[List[ReasoningStep]]	None	All reasoning steps taken
reasoning_messages	Optional[List[Message]]	None	Messages from the reasoning process
run_id	Optional[str]	None	ID of the current run
agent_id	Optional[str]	None	ID of the reasoning agent
created_at	int	Current timestamp	Unix timestamp of event creation

Agent Configuration for Reasoning

Reasoning Agent Parameters

Parameters for configuring Reasoning Agents (reasoning=True):

Parameter	Type	Default	Description
reasoning	bool	False	Enable reasoning agent mode
reasoning_model	Optional[Model]	None	Separate model for reasoning (if different from main model)
reasoning_agent	Optional[Agent]	None	Custom reasoning agent instance
reasoning_min_steps	int	1	Minimum number of reasoning steps
reasoning_max_steps	int	10	Maximum number of reasoning steps

Display Parameters

Parameters for showing reasoning during execution:

Parameter	Type	Default	Description
show_full_reasoning	bool	False	Display complete reasoning process in output
stream_events	bool	False	Stream each reasoning step in real-time

See Also

• Reasoning Overview - Introduction to reasoning approaches
• Reasoning Agents Guide - Using Reasoning Agents
• Reasoning Tools Guide - Using Reasoning Tools
• Reasoning Models Guide - Using native reasoning models