AI SDK Kotlin/ai.torad.aisdk/StepSettings

StepSettings

Returned from prepareStep. Any non-null field overrides the agent default / prepareCall value for this step only.

Sampler params here override the per-invocation values from AgentSettings for one step — same vocabulary as v6's tool-loop-agent-settings.ts's prepareStep return. Useful when (say) step 1 is a deterministic plan (temperature = 0) and the following synthesis steps are warmer.

Since

0.3.0-beta01

Properties

activeTools

val activeTools: List<String>?

experimental_context

val experimental_context: TContext?

Per-step typed-context override. Mirrors v6's prepareStep.experimental_context (per historical parity gap #16). When set, subsequent steps see this value as their ToolExecutionContext.context — useful for mid-loop context evolution (e.g., RAG augmentation after a tool result).

frequencyPenalty

val frequencyPenalty: Float?

maxOutputTokens

val maxOutputTokens: Int?

maxRetries

val maxRetries: Int?

messages

val messages: List<ModelMessage>?

model

val model: LanguageModel?

presencePenalty

val presencePenalty: Float?