AgentTrajectory

Models & Types · AiDotNet.Agentic.SelfImproving

A captured record of one agent run — the structured “trajectory” the self-improving layer learns from. It pairs what the agent did (the full message transcript, the final answer, step count, token usage) with an optional quality Reward assigned later by an evaluator.

For Beginners

Think of this as the flight recorder for an agent: it saves exactly what was said and done on a run, plus (once graded) how good the outcome was. Collect many of these and the system can learn what works and get better over time.

How It Works

Trajectories are the raw material for every self-improvement mechanism: continuous evaluation scores them, learned routers/tool policies train on them, prompt optimizers measure prompts against them, and reward-filtered trajectories become fine-tuning (LoRA) data. Capturing a run is non-invasive — a TracingAgent wraps any agent and records each run without changing its behavior.

Constructors

Constructor Summary
AgentTrajectory(string id, string agentName, IReadOnlyList<ChatMessage> messages, string finalText, int iterations, ChatUsage usage, double? reward, IReadOnlyDictionary<string, string> metadata) Initializes a new trajectory.

Properties

Property Summary
ChatUsage Usage { get; } Gets aggregate token usage, or null when not reported.
IReadOnlyDictionary<string, string> Metadata { get; } Gets optional key/value metadata, or null when none was supplied.
IReadOnlyList<ChatMessage> Messages { get; } Gets the full conversation transcript the run produced.
double? Reward { get; set; } Gets or sets the quality score for this trajectory, assigned by an evaluator (higher is better), or null while ungraded.
int Iterations { get; } Gets the number of model calls the run took.
string AgentName { get; } Gets the name of the agent that produced the run.
string FinalText { get; } Gets the agent’s final answer.
string Id { get; } Gets the unique id for this trajectory.