Package com.tnsai.reasoning

Interface ReasoningStrategyHandle

public interface ReasoningStrategyHandle

SPI handle for reasoning strategy integration.

This interface defines the reasoning contract that Agent uses. Concrete implementations live in tnsai-intelligence and are discovered via SpiLoader.

Supported strategies include:

• ReAct: Reason → Act → Observe loop
• Tree of Thoughts: Multi-path exploration with pruning
• Graph of Thoughts: Graph-based reasoning with merging
• Self-Consistency: Multiple paths with consensus voting

Usage Example

 Agent agent = AgentBuilder.create()
     .role(new MyRole())
     .reasoningStrategy("react")
     .build();

 // Agent uses ReAct loop instead of simple LLM call
 String answer = agent.chat("Solve this problem step by step");

 // Access reasoning details
 agent.getLastReasoningResult().ifPresent(result -> {
     System.out.println("Steps: " + result.metadata().get("steps"));
     System.out.println("LLM calls: " + result.metadata().get("llmCalls"));
 });
 

Nested Class Summary

Nested Classes

Modifier and Type	Interface	Description
static interface	ReasoningStrategyHandle.Factory	Factory for creating ReasoningStrategyHandle instances by name.
static final record	ReasoningStrategyHandle.ReasoningContext	Context provided to the reasoning strategy by the Agent.
static final record	ReasoningStrategyHandle.ReasoningResult	Result of a reasoning operation.

Method Summary

Modifier and Type	Method	Description
String	name()	Returns the strategy name (e.g., "react", "tot", "got", "self-consistency").
ReasoningStrategyHandle.ReasoningResult	reason(String message, ReasoningStrategyHandle.ReasoningContext context)	Performs reasoning on a message using the strategy's algorithm.

Method Details

reason

ReasoningStrategyHandle.ReasoningResult reason(String message,
 ReasoningStrategyHandle.ReasoningContext context)

Performs reasoning on a message using the strategy's algorithm.

Parameters:

message - user message to reason about

context - reasoning context providing LLM, tools, and history

Returns:

result containing the answer and strategy-specific metadata

name

String name()

Returns the strategy name (e.g., "react", "tot", "got", "self-consistency").

Returns:

strategy identifier