koog-agents-optimization/ai.koog.agents.optimization.consumption/LLMConsumption

LLMConsumption

@Serializable(with = LLMConsumptionSerializer::class)

Base type for an amount of LLM consumption recorded during an optimization run.

This is an abstract class (not sealed) so that providers outside the library can define their own consumption flavors. Each flavor pairs with an LLMConsumptionUnit and registers a decoder via LLMConsumptionType (discovered through LLMConsumptionRegistry); serialization is handled uniformly by LLMConsumptionSerializer.

Inheritors

LiteLLMTokenConsumption

Constructors

LLMConsumption

constructor()

Properties

total

abstract val total: Double

The single scalar total used for comparisons (e.g. against a spend cap).

unit

abstract val unit: LLMConsumptionUnit

The unit this consumption is measured in.

Functions

plus

abstract operator fun plus(consumption: LLMConsumption): LLMConsumption

Adds two consumptions of the same flavor. Implementations reject mismatched flavors.

toPrettyJson

abstract fun toPrettyJson(): JsonObject

The flavor-specific fields as pretty JSON (without the top-level "unit" field).

toPrettyString

abstract fun toPrettyString(): String

Human-readable form, typically with input/output breakdown where applicable.

toPrettyTotal

fun toPrettyTotal(): String

Total-only short form, e.g. "330,329.00 tokens" / "150.00 credits". Used where the input/output breakdown would be misleading or off-topic — most notably the spend-limit abort message, where there is no such thing as a separate input/output cap (cap is a single total).