koog-agents-optimization/ai.koog.agents.optimization.optimizers

Package-level declarations

Types

interface AgentOptimizer<Input, Output, InputLabel>

Common contract for agent optimizers: algorithms that improve a GraphAIAgent by learning from execution trajectories collected over a training dataset.

TrainSet

typealias TrainSet<AgentInput, ItemLabel> = List<TrainSetItem<AgentInput, ItemLabel>>

A training dataset: an ordered list of TrainSetItems an optimizer runs the agent on.

TrainSetItem

data class TrainSetItem<AgentInput, ItemLabel>(val userQuery: AgentInput, val itemLabel: ItemLabel)

An optimizer requires dataset items to run an agent on them and collect execution trajectories. DatasetItems consist of a userQuery, which is being passed to the Agent, and a label, which is the golden answer. The one using an optimizer is in charge of providing an optimizer with a relevant metric that can evaluate the produced result over the golden label.