koog-spring-ai-starter-model-chat/ai.koog.spring.ai.chat/SpringAiLLMClient

SpringAiLLMClient

class SpringAiLLMClient(chatModel: ChatModel, provider: LLMProvider, clock: Clock, dispatcher: CoroutineDispatcher, chatOptionsCustomizer: ChatOptionsCustomizer, moderationModel: ModerationModel?) : LLMClient(source)

An LLMClient implementation that delegates to a Spring AI ChatModel.

This adapter allows Koog agents to use any Spring AI chat model provider (Anthropic, OpenAI, Google, Ollama, etc.) as their underlying LLM backend.

Tool execution is always owned by the Koog agent framework. Spring AI receives only tool definitions (via org.springframework.ai.tool.ToolCallback with a throwing call()) and internalToolExecutionEnabled=false, so Spring never attempts to execute tools.

Parameters

chatModel

the Spring AI chat model to delegate to

provider

the LLMProvider to report for this client

clock

the clock used for creating response metadata timestamps

dispatcher

the CoroutineDispatcher used for blocking model calls

chatOptionsCustomizer

optional customizer for provider-specific ChatOptions tuning

moderationModel

optional Spring AI ModerationModel for content moderation; if null, moderate throws UnsupportedOperationException

Constructors

SpringAiLLMClient

constructor(chatModel: ChatModel, provider: LLMProvider, clock: Clock, dispatcher: CoroutineDispatcher, chatOptionsCustomizer: ChatOptionsCustomizer, moderationModel: ModerationModel?)

Types

Builder

class Builder

A Java-friendly builder for SpringAiLLMClient.

Companion

object Companion

Java-friendly builder access.

Properties

clientName

open val clientName: String

Functions

open override fun close()

execute

@JvmOverloads

fun execute(prompt: Prompt, model: LLModel, tools: List<ToolDescriptor>, executorService: ExecutorService?): List<Message.Response>

open suspend override fun execute(prompt: Prompt, model: LLModel, tools: List<ToolDescriptor>): List<Message.Response>

executeMultipleChoices

@JvmOverloads

fun executeMultipleChoices(prompt: Prompt, model: LLModel, tools: List<ToolDescriptor>, executorService: ExecutorService?): List<LLMChoice>

open suspend fun executeMultipleChoices(prompt: Prompt, model: LLModel, tools: List<ToolDescriptor>): List<LLMChoice>

executeStreaming

open fun executeStreaming(prompt: Prompt, model: LLModel): Flow<StreamFrame>

open override fun executeStreaming(prompt: Prompt, model: LLModel, tools: List<ToolDescriptor>): Flow<StreamFrame>

Streams LLM responses by subscribing to ChatModel.stream and converting each chunk into Koog StreamFrame events.

executeStreamingWithPublisher

@JvmOverloads

fun executeStreamingWithPublisher(prompt: Prompt, model: LLModel, tools: List<ToolDescriptor>): Flow.Publisher<StreamFrame>

getBasicJsonSchemaGenerator

open fun getBasicJsonSchemaGenerator(): BasicJsonSchemaGenerator

getStandardJsonSchemaGenerator

open fun getStandardJsonSchemaGenerator(): StandardJsonSchemaGenerator

llmProvider

open override fun llmProvider(): LLMProvider

models

@JvmOverloads

fun models(executorService: ExecutorService?): List<LLModel>

open suspend override fun models(): List<LLModel>

Returns the list with one model based on the configured LLMProvider and ChatModel without capabilities or parameters.

moderate

@JvmOverloads

fun moderate(prompt: Prompt, model: LLModel, executorService: ExecutorService?): ModerationResult

open suspend override fun moderate(prompt: Prompt, model: LLModel): ModerationResult