AI SDK Kotlin/ai.torad.aisdk/LanguageModelCallParams

LanguageModelCallParams

data class LanguageModelCallParams(val messages: List<ModelMessage>, val tools: List<LanguageModelTool> = emptyList(), val toolChoice: ToolChoice = ToolChoice.Auto, val temperature: Float? = null, val topP: Float? = null, val topK: Int? = null, val maxOutputTokens: Int? = null, val stopSequences: List<String> = emptyList(), val seed: Int? = null, val providerOptions: Map<String, JsonElement> = emptyMap(), val abortSignal: AbortSignal = AbortSignalNever, val presencePenalty: Float? = null, val frequencyPenalty: Float? = null, val responseFormat: ResponseFormat = ResponseFormat.Text, val headers: Map<String, String> = emptyMap())

Single parameter envelope for both generate and stream so middleware can wrap both the same way and wrapLanguageModel only needs one pass-through shape.

Constructors

constructor(messages: List<ModelMessage>, tools: List<LanguageModelTool> = emptyList(), toolChoice: ToolChoice = ToolChoice.Auto, temperature: Float? = null, topP: Float? = null, topK: Int? = null, maxOutputTokens: Int? = null, stopSequences: List<String> = emptyList(), seed: Int? = null, providerOptions: Map<String, JsonElement> = emptyMap(), abortSignal: AbortSignal = AbortSignalNever, presencePenalty: Float? = null, frequencyPenalty: Float? = null, responseFormat: ResponseFormat = ResponseFormat.Text, headers: Map<String, String> = emptyMap())

Properties

abortSignal

val abortSignal: AbortSignal

frequencyPenalty

val frequencyPenalty: Float?

Penalty proportional to how often a token has appeared. Mirrors v6's CallSettings. Null = provider default.

headers

val headers: Map<String, String>

maxOutputTokens

val maxOutputTokens: Int?

messages

val messages: List<ModelMessage>

presencePenalty

val presencePenalty: Float?

Penalty for repeating tokens that already appeared in the response, regardless of frequency. Mirrors v6's CallSettings (per historical parity gap #3 closure). Null = provider default.

providerOptions

val providerOptions: Map<String, JsonElement>

responseFormat

val responseFormat: ResponseFormat

Wire-level shape constraint. Mirrors v6's LanguageModelV3CallOptions.responseFormat (per historical parity gap #20). Providers that support constrained decoding honor it; others ignore. Default ResponseFormat.Text = no constraint.

seed

val seed: Int?

stopSequences

val stopSequences: List<String>

temperature

val temperature: Float?

toolChoice

val toolChoice: ToolChoice

tools

val tools: List<LanguageModelTool>

topK

val topK: Int?

topP

val topP: Float?