Table of Contents

Class ChainOfThoughtHandler

Namespace
LMKit.Agents.Planning
Assembly
LM-Kit.NET.dll

Planning handler that implements Chain-of-Thought (CoT) prompting.

Encourages the model to reason through problems step-by-step before providing a final answer. Improves performance on arithmetic, logic, and multi-step reasoning tasks. 

public sealed class ChainOfThoughtHandler : PlanningHandlerBase, IPlanningHandler
Inheritance
ChainOfThoughtHandler
Implements
Inherited Members

Examples

Example: Create an agent with Chain-of-Thought planning

using LMKit.Model;
using LMKit.Agents;
using System;
using System.Threading.Tasks;

LM model = LM.LoadFromModelID("llama-3.2-3b");

// Build agent with Chain-of-Thought planning
Agent agent = new AgentBuilder()
    .WithModel(model)
    .WithPersona("Math Tutor")
    .WithInstruction("Solve problems by showing your work step by step.")
    .WithPlanning(PlanningStrategy.ChainOfThought)
    .Build();

// Execute - model will reason through the problem
using var executor = new AgentExecutor();
var result = await executor.ExecuteAsync(
    agent, 
    "If a train travels 120 miles in 2 hours, what is its average speed?");

Console.WriteLine($"Answer: {result.Content}");
Console.WriteLine($"Reasoning:\n{executor.PlanningContext?.GetReasoningTrace()}");

Example: Complex reasoning with Chain-of-Thought

using LMKit.Model;
using LMKit.Agents;
using System;
using System.Threading.Tasks;

LM model = LM.LoadFromModelID("llama-3.2-3b");

Agent agent = new AgentBuilder()
    .WithModel(model)
    .WithPersona("Logic Expert")
    .WithPlanning(PlanningStrategy.ChainOfThought)
    .Build();

using var executor = new AgentExecutor();
var result = await executor.ExecuteAsync(agent, @"
    All roses are flowers. 
    Some flowers fade quickly. 
    Can we conclude that some roses fade quickly?");

Console.WriteLine(result.Content);

Remarks

How It Works
The handler wraps user input with instructions to "think step by step" and extracts both the reasoning process and the final answer from the output.

Output Format
The handler looks for a "Final Answer:" marker to separate reasoning from the answer. If not found, the entire output is treated as the answer.

Constructors

ChainOfThoughtHandler()

Initializes a new instance of the ChainOfThoughtHandler class.

Fields

DefaultInstruction

The default instruction added to encourage step-by-step reasoning.

FinalAnswerMarker

The marker indicating the final answer.

Properties

Instance

Gets the singleton instance with default settings.

ReasoningInstruction

Gets or sets the instruction to append for step-by-step reasoning.

RequireFinalAnswerMarker

Gets or sets a value indicating whether to require a "Final Answer:" marker.

When true, the handler expects the model to explicitly mark its final answer. When false, the entire output is used if no marker is found.

Strategy

Gets the planning strategy.

Methods

PrepareInput(PlanningContext, string)

Prepares the input by appending the step-by-step reasoning instruction.

ProcessOutput(PlanningContext, string)

Processes the model output by extracting reasoning and final answer.

WithInstruction(string)

Creates a handler with custom reasoning instruction.

WithRequiredMarker()

Creates a handler that requires explicit "Final Answer:" markers.