agents-test/ai.koog.agents.testing.client/CapturingLLMClient

CapturingLLMClient

class CapturingLLMClient @JvmOverloads constructor(executeResponse: Message.Assistant? = null, streamingChunks: List<StreamFrame> = emptyList(), choices: LLMChoice? = null, moderationResult: ModerationResult = ModerationResult(isHarmful = false, categories = emptyMap()), embedResult: List<Double> = emptyList(), batchEmbedResult: List<List<Double>> = emptyList(), llmProvider: LLMProvider = LLMProvider.OpenAI) : LLMClient(source)

A test double implementation of LLMClient that captures the last inputs provided to each API and returns predefined responses. This is useful in unit and integration tests to assert that a component under test interacts with an LLM client as expected without making real network calls.

Constructor parameters allow you to predefine what each method should return.

Constructors

CapturingLLMClient

@JvmOverloads

constructor(executeResponse: Message.Assistant? = null, streamingChunks: List<StreamFrame> = emptyList(), choices: LLMChoice? = null, moderationResult: ModerationResult = ModerationResult(isHarmful = false, categories = emptyMap()), embedResult: List<Double> = emptyList(), batchEmbedResult: List<List<Double>> = emptyList(), llmProvider: LLMProvider = LLMProvider.OpenAI)

Properties

clientName

open val clientName: String

lastBatchEmbeddingInput

var lastBatchEmbeddingInput: List<String>?

The last batched input passed to embed, or null if it hasn't been called yet.

lastBatchEmbeddingModel

var lastBatchEmbeddingModel: LLModel?

The last LLModel passed to embed, or null if it hasn't been called yet.

lastChoicesModel

var lastChoicesModel: LLModel?

The last LLModel passed to executeMultipleChoices, or null if it hasn't been called yet.

lastChoicesPrompt

var lastChoicesPrompt: Prompt?

The last Prompt passed to executeMultipleChoices, or null if it hasn't been called yet.

lastChoicesTools

var lastChoicesTools: List<ToolDescriptor>?

The last list of tools passed to executeMultipleChoices, or null if it hasn't been called yet.

lastEmbeddingModel

var lastEmbeddingModel: LLModel?

The last LLModel passed to embed, or null if it hasn't been called yet.

lastEmbeddingText

var lastEmbeddingText: String?

The last text passed to embed, or null if it hasn't been called yet.

lastExecutedModel

var lastExecutedModel: LLModel?

The last LLModel passed to execute, or null if it hasn't been called yet.

lastExecutedPrompt

var lastExecutedPrompt: Prompt?

The last Prompt passed to execute, or null if it hasn't been called yet.

lastExecutedTools

var lastExecutedTools: List<ToolDescriptor>?

The last list of tools passed to execute, or null if it hasn't been called yet.

lastModerationModel

var lastModerationModel: LLModel?

The last LLModel passed to moderate, or null if it hasn't been called yet.

lastModerationPrompt

var lastModerationPrompt: Prompt?

The last Prompt passed to moderate, or null if it hasn't been called yet.

lastStreamingModel

var lastStreamingModel: LLModel?

The last LLModel passed to executeStreaming, or null if it hasn't been called yet.

lastStreamingPrompt

var lastStreamingPrompt: Prompt?

The last Prompt passed to executeStreaming, or null if it hasn't been called yet.

Functions

open override fun close()

embed

open suspend override fun embed(text: String, model: LLModel): List<Double>

Simulates an embedding call. Captures input parameters and returns the predefined embedResult.

open suspend override fun embed(inputs: List<String>, model: LLModel): List<List<Double>>

Simulates a batch embedding call. Captures input parameters and returns the predefined batchEmbedResult.

execute

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

Simulates a non-streaming LLM execution. Captures input parameters and returns the predefined executeResponse.

executeMultipleChoices

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

Simulates an LLM call that returns multiple choices. Captures input parameters and returns the predefined choices.

executeStreaming

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

Simulates a streaming LLM execution. Captures input parameters and emits the predefined streamingChunks.

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

getBasicJsonSchemaGenerator

open fun getBasicJsonSchemaGenerator(): BasicJsonSchemaGenerator

getStandardJsonSchemaGenerator

open fun getStandardJsonSchemaGenerator(): StandardJsonSchemaGenerator

llmProvider

open override fun llmProvider(): LLMProvider

models

open suspend fun models(): List<LLModel>

moderate

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

Simulates a content moderation call. Captures input parameters and returns the predefined moderationResult.