Class PlanningContext

Namespace: LMKit.Agents.Planning

Assembly: LM-Kit.NET.dll

Provides context and state for planning strategy execution.

The context is passed to IPlanningHandler methods and contains the agent configuration, execution state, and accumulated results.

public sealed class PlanningContext

Inheritance: object

PlanningContext

Inherited Members: object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Examples

Example: Access planning context during agent execution

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

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

// Build agent with ReAct planning
Agent agent = new AgentBuilder()
    .WithModel(model)
    .WithPersona("Research Assistant")
    .WithPlanning(PlanningStrategy.ReAct)
    .Build();

using var executor = new AgentExecutor();

// Execute and access planning context
var result = await executor.ExecuteAsync(agent, "Analyze this problem step by step.");

// Access planning context after execution
PlanningContext context = executor.PlanningContext;
if (context != null)
{
    Console.WriteLine($"Iterations: {context.Iteration}");
    Console.WriteLine($"Steps: {context.Steps.Count}");
    Console.WriteLine($"Reasoning trace:\n{context.GetReasoningTrace()}");
}

Example: Store custom state in planning context

using LMKit.Agents.Planning;
using System;

// During custom planning handler implementation
void ProcessStep(PlanningContext context)
{
    // Store intermediate results
    context.SetState("processedItems", 5);
    context.SetState("confidence", 0.95);

    // Retrieve state later
    int items = context.GetState<int>("processedItems");
    double confidence = context.GetState<double>("confidence", 0.0);

    // Append to reasoning trace
    context.AppendReasoningTrace($"Processed {items} items with {confidence:P0} confidence");

    // Check iteration limits
    if (context.MaxIterationsReached)
    {
        Console.WriteLine("Max iterations reached, finalizing...");
    }
}

Remarks

Lifecycle
A new context is created for each agent execution. It accumulates state across planning iterations and is disposed when execution completes.

Thread Safety
This class is not thread-safe and should only be accessed from a single execution thread.

Properties

Agent: Gets the agent being executed.

CurrentAction: Gets or sets the current action being taken.

CurrentThought: Gets or sets the current thought or reasoning step.

FinalAnswer: Gets or sets the final answer when planning is complete.

Iteration: Gets the current iteration number (1-based).

LastObservation: Gets or sets the most recent observation (e.g., tool result).

LastOutput: Gets or sets the most recent model output.

MaxIterations: Gets the maximum allowed iterations.

MaxIterationsReached: Gets a value indicating whether the maximum iterations have been reached.

OriginalInput: Gets the original user input.

Steps: Gets the list of completed planning steps.

ToolCalls: Gets the tool calls made during the current planning cycle.

Methods

AppendReasoningTrace(string): Appends content to the reasoning trace.

GetReasoningTrace(): Gets the accumulated reasoning trace.

GetState<T>(string, T): Retrieves a value from the context state.

HasState(string): Checks if a state key exists.

SetState(string, object): Stores a value in the context state.

Table of Contents

Class PlanningContext

Examples

Remarks

Properties

Methods