AI SDK Kotlin/ai.torad.aisdk/OnStepStartEvent

OnStepStartEvent

data class OnStepStartEvent(val stepNumber: Int, val messages: List<ModelMessage>, val request: LanguageModelCallParams? = null, val priorSteps: List<StepResult> = emptyList())

Fired before each loop step's model call. Per historical parity gap #35, the payload was extended from (stepNumber, messages) to carry the full prepared request (post prepareCall + prepareStep overrides) plus accumulated priorSteps so observers can:

Inspect the EXACT params being sent (system prompt, tool subset, sampler overrides) — useful for telemetry and reproduction.
Read prior-step output without re-walking the StepResult list from a separate accumulator.

Both new fields default to null / empty so existing observers keep working unchanged. Loop-side population is staged in as a follow-up; the type surface is in place now.

Constructors

OnStepStartEvent

constructor(stepNumber: Int, messages: List<ModelMessage>, request: LanguageModelCallParams? = null, priorSteps: List<StepResult> = emptyList())

Properties

messages

val messages: List<ModelMessage>

priorSteps

val priorSteps: List<StepResult>

Accumulated prior-step results — priorSteps[i] is step i+1's outcome. Empty on step 1.

request

val request: LanguageModelCallParams?

The prepared call params for this step. Null when the loop hasn't yet wired the field (back-compat). When set, includes the resolved system prompt, tool subset, sampler params, etc.

stepNumber

val stepNumber: Int