Krosai

Krosai is a Kotlin Multiplatform library that provides a simple and easy-to-use API for interacting with AI models. It supports multiple AI models, including OpenAI and more.

English | 简体中文

Features

• Multiplatform: Krosai is a Kotlin Multiplatform library that supports JVM, Android, iOS and WASM.
• Simple API: Krosai combined with Kotlin DSL provides easy-to-use APIs for interacting with AI models.
• Multiple AI Models: Krosai supports multiple AI models, including OpenAI.

Supported

API

• Chat
  • Call
  • Stream
  • Prompt
  • FunctionCall
  • Enhancer
    • ChatMemory
• Embedding
• Image
• AudioTranscription(TODO)
• Speech(TODO)

AI Models

• OpenAI

Vector Store

• MongoDB(TODO)

Usage Example

The following example uses the OpenAI model, and the other models are used in a similar way.

// 1. create a `ModelFactoryContext` 
val context = buildModelFactoryContext {
    // Using OpenAI Model
    factory(OpenAI) {
        baseUrl = "https://api.openai.com"
        apiKey = "YOUR_API_KEY"
    }
}
// 2. get a `ModelFactory` from the it
val modelFactory = context[OpenAI]

• Chat
  • Prompt

    // create a `ChatClient` from the `ModelFactory`
    val chatClient = modelFactory.createChatClient{
      // set the Default Request of the ChatClient
      // for example, set the `systemText`
      systemText {"Please answer in ${get("language")}"}
    }
    // call or stream the `ChatClient` with the request
    chatClient.call{
      // set the `userText` for the request
      userText { "Please call me Kamo." }
      system { 
          // Fill 'language' with "Japanese" in systemText"
          "language" to "Japanese"
      }
    }.let { response ->
         // print the response content
         println(response.content)
      }

  • Enhancer

    val chatClient = context.createChatClient {
        enhancers {
            // add the `ChatMemoryEnhancer` to the Default Request of the `ChatClient`
            // `InMemoryMessageStore` is used to store the messages in memory
            +MessageChatMemoryEnhancer(InMemoryMessageStore())
        }
    }
    chatClient.call {
        userText { "Please call me Kamo." }
    }
    
    chatClient.call {
        userText { "Please answer in Japanese" }
    }.let {
        // respond in japanese
        println("content: ${it.content}")
        // the response contains "Kamo", because it is stored in Memory
        assert(it.content.contains("Kamo"))
    }

  • FunctionCall

    Define a FunctionCall and use it in the Request.

    @SerialDescription("get date time API Request")
    @Serializable
    data class Request(
        @SerialName("location") 
        @SerialDescription("location")
        val location: String
    )
    val dateTimeFun = buildFunctionCall {
        name = "date_time"
        description = "Get the current date by location"
        withCall { req: Request ->
            "the current time in ${req.location} is ${LocalDateTime.now()}"
        }
    }

    chatClient.call {
       userText { "What time is it in New York?" }
       functions {
           +dateTimeFun
       }
    .let {
      // print the response content containing the current time in New York
       println(it.content)
    }