koog-sauce：连接 Koog 与 Spring AI、LangChain4j 的集成库

Koog-sauce is a missing ingredient that connects Koog with other frameworks.

Features

• Spring AI Integration - Provides a Koog's SpringAiLLMClient, which uses Spring AI's ChatClient under the hood.
• LangChain4j Integration - Offers Langchain4jLLMClient for seamless integration with LangChain4j, supporting both standard and streaming interactions.
• AI Agent Builder - Simplifies creation and configuration of AI agents with a fluent builder pattern, making it easy to set up complex agent behaviors.

Requirements

• JDK 17 or higher
• Gradle 8.14.1 or higher

Getting Started

Add Dependency

Add the dependency to your build.gradle.kts file:

dependencies {
    // Core library
    implementation("me.kpavlov:koog-sauce:[LATEST]")

    // For Spring AI integration
    implementation("me.kpavlov:koog-sauce-spring-ai:[LATEST]")
    implementation("org.springframework.ai:spring-ai-openai:1.0.0")

    // For LangChain4j integration
    implementation("me.kpavlov:koog-sauce-langchain4j:[LATEST]")
    implementation("dev.langchain4j:langchain4j:0.24.0")
    implementation("dev.langchain4j:langchain4j-open-ai:0.24.0")

    // Koog library
    implementation("ai.koog:koog:0.4.0") // or newer
}

Build the Project

./gradlew build

Or using the Makefile:

make build

Usage

Spring AI Integration

// Create Spring AI ChatClient
val chatClient = org.springframework.ai.chat.client.ChatClient.builder(
    org.springframework.ai.openai.OpenAiChatModel
        .builder()
        .openAiApi(
            org.springframework.ai.openai.api.OpenAiApi
                .builder()
                .apiKey("your-api-key")
                .build(),
        ).build(),
).build()

// Create SpringAiLLMClient
val llmClient = me.kpavlov.koog.sauce.spring.ai.chat.SpringAiLLMClient(chatClient)

// Build a prompt using Koog DSL
val prompt = ai.koog.prompt.dsl.Prompt.build("myPrompt") {
    system("You are a helpful assistant")
    user("Tell me about Kotlin Multiplatform")
}

// Define the model to use
val model = ai.koog.prompt.llm.LLModel(
    ai.koog.prompt.llm.LLMProvider.OpenAI, 
    "gpt-4.1-nano", 
    listOf(ai.koog.prompt.llm.LLMCapability.Completion),
    100500
)

// Execute the prompt
suspend fun executePrompt() {
    val responses = llmClient.execute(prompt, model)

    // Process the response
    val response = responses.first()
    println("Response: ${response.content}")
}

See the complete example.

LangChain4j Integration

// Create LangChain4j ChatModel
val chatModel = dev.langchain4j.model.openai.OpenAiChatModel.builder()
    .apiKey("your-api-key")
    .modelName("gpt-4")
    .build()

// Create Langchain4jLLMClient
val llmClient = me.kpavlov.koog.sauce.langchain4j.Langchain4jLLMClient(chatModel = chatModel)

// Build a prompt using Koog DSL
val prompt = ai.koog.prompt.dsl.Prompt.build("myPrompt") {
    system("You are a helpful assistant")
    user("Tell me about LangChain4j")
}

// Define the model to use
val model = ai.koog.prompt.llm.LLModel(
    ai.koog.prompt.llm.LLMProvider.OpenAI, 
    "gpt-4", 
    listOf(ai.koog.prompt.llm.LLMCapability.Completion),
    100500,
)

// Execute the prompt
suspend fun executePrompt() {
    val responses = llmClient.execute(prompt, model)
    println("Response: ${responses.first().content}")
}

See the complete example.

AI Agent Builder

// Create a prompt executor
val promptExecutor = ai.koog.prompt.executor.model.PromptExecutor(
    llmClient = llmClient,
    defaultModel = model
)

// Create an AI agent using the builder
val agent = me.kpavlov.koog.sauce.agents.core.agent.AIAgent<String, String> {
    this.promptExecutor = promptExecutor
    this.strategy = YourCustomStrategy() // Implement AIAgentStrategy
    this.agentConfig = YourAgentConfig() // Implement AIAgentConfigBase
    this.toolRegistry = ToolRegistry.builder()
        .registerTool(YourCustomTool()) // Add your tools
        .build()
}

// Use the agent
suspend fun executeAgent() {
    val result = agent.execute("Tell me about Kotlin")
    println("Agent result: $result")
}

These examples demonstrate how to:

1. Integrate with Spring AI and LangChain4j
2. Build AI agents with the fluent builder pattern
3. Configure different components like prompt executors, strategies, and tools

Development

Available Make Commands

• make build: Build the project
• make test: Run tests
• make clean: Clean the project
• make publish: Publish to Maven Local
• make doc: Generate KDoc documentation
• make help: Show help message

Running Tests

./gradlew test

Or using the Makefile:

make test

Integration Tests

The project includes comprehensive tests for both Spring AI and LangChain4j integrations:

• Spring AI Tests: SpringOpenAiTest demonstrates how to use and test the Spring AI integration with OpenAI models.
• LangChain4j Tests:
  • Langchain4jLLMClientTest tests the standard LangChain4j LLM client functionality.
  • StreamingLangchain4jLLMClientTest tests the streaming capabilities of the LangChain4j LLM client.

These tests serve as additional examples of how to use the integrations in your own projects.

Generating Documentation

./gradlew dokkaGeneratePublicationHtml

Or using the Makefile:

make doc

The documentation is automatically generated and published to GitHub Pages when changes are pushed to the main branch. You can access the latest documentation at:

https://kpavlov.github.io/koog-sauce/

License

This project is licensed under the MIT License - see the LICENSE file for details.