DeepSeek Kotlin SDK/org.oremif.deepseek.models.responses/ResponseParams

ResponseParams

Everything about a Responses API call except the conversation itself.

The Responses API is stateless, so the same parameters are reused turn after turn while only the input grows:

val params = responseParams { instructions = "You are a helpful assistant." }
val history = mutableListOf<ResponseItem>(MessageItem(ResponseRole.USER, "Hi"))
val first = client.createResponse(params, history)
history += first.output

Every constraint the API documents is checked on construction, so an invalid combination fails here rather than as a 400 from the server.

Types

Builder

class Builder

Builder for creating ResponseParams with standard (non-streaming) configuration.

StreamBuilder

class StreamBuilder

Builder for creating ResponseParams configured for streaming responses.

Properties

instructions

val instructions: String?

System-level instruction, inserted as the first system message of the model's context. Either this or an input is required.

maxOutputTokens

val maxOutputTokens: Int?

Ceiling on the tokens generated for the response, counting both the visible answer and the reasoning. At least 1.

model

val model: ChatModel

The DeepSeek model to use; defaults to ChatModel.DEEPSEEK_FLASH.

reasoning

val reasoning: ReasoningConfig?

Thinking-mode configuration; see ReasoningConfig.

stream

val stream: Boolean?

Whether the response is streamed as semantic server-sent events.

temperature

val temperature: Double?

Sampling temperature between 0.0 and 2.0; has no effect in thinking mode.

text

val text: TextConfig?

Shape of the answer, including JSON-Schema-constrained output; see TextConfig.

toolChoice

val toolChoice: ResponseToolChoice?

Which tool, if any, the model must call.

tools

val tools: List<ResponseTool>?

Tools the model may use — functions, and the server-side web search.

topLogprobs

val topLogprobs: Int?

How many most likely tokens to report at each position; 0 to 20.

topP

val topP: Double?

Nucleus sampling mass between 0.0 and 1.0; has no effect in thinking mode.

user

val user: String?

Custom identifier of the end user behind the request; 1 to 512 characters from [a-zA-Z0-9\-_]. Must not carry personal data. The chat endpoint spells the same field user_id.

Functions

copy

fun copy(model: ChatModel = this.model, instructions: String? = this.instructions, reasoning: ReasoningConfig? = this.reasoning, maxOutputTokens: Int? = this.maxOutputTokens, stream: Boolean? = this.stream, temperature: Double? = this.temperature, topP: Double? = this.topP, text: TextConfig? = this.text, tools: List<ResponseTool>? = this.tools, toolChoice: ResponseToolChoice? = this.toolChoice, topLogprobs: Int? = this.topLogprobs, user: String? = this.user): ResponseParams

Creates a copy of these parameters with optional changes to specific properties.

createRequest

fun createRequest(): ResponseRequest

Creates a ResponseRequest with no input, driven by instructions alone.

fun createRequest(input: String): ResponseRequest

Creates a ResponseRequest carrying a single user message.

fun createRequest(input: List<ResponseItem>): ResponseRequest

Creates a ResponseRequest carrying a conversation.

fun createRequest(input: ResponseInput?): ResponseRequest

Creates a ResponseRequest from an already-shaped input.

equals