Release Kotlin package with DSL builders and multiplatform support

[gad2103]

My company uses Kotlin server side and it would be nice to have Kotlin DSL builders for the all the objects herein. Happy to help contribute on this if you are looking for some contributors.

[TomerAberbach]

Hey @gad2103!

We actually have plans to eventually release a Kotlin specific package (openai-kotlin) where the public API is specifically tailored for Kotlin. For example, openai-kotlin would have:

• Nullable return types instead of optionals
• Computed properties instead of getter methods
• Kotlin-friendly builders :)

For the builders, we were thinking something like this:

val params = ChatCompletionCreateParams {
  messages(/* ... */) 
  // OR
  userMessage {
    // ...
  }
  assistantMessage {
    // ...
  }
  // More messages...

  model(ChatModel.O1)
  // More properties...
}

Does that look like roughly what you wanted?

[nathanfallet]

Any ETA for the kotlin specific package? (it's based on the java one, only adding extensions)

[TomerAberbach]

Any ETA for the kotlin specific package? (it's based on the java one, only adding extensions)

We can't make any firm commitments right now, but we'd hope to ship a Kotlin package within a month or two of GA'ing this Java one.

[devcrocod]

Hi!
Do you have any plans for Kotlin Multiplatform?

[TomerAberbach]

Yes, the Kotlin package would support multiplatform

[Taewan-P]

Hello! Any further updates on this? Having Kotlin package would be really great!

[TomerAberbach]

We still don't have a firm commitments (see #70 (comment)), but we've been actively working on getting this Java package to GA status, which will help the Kotlin package

[agbragin]

Hi! Any updates on this? It looks like openai-java is out of beta now.

[nathanfallet]

Actually we're switching to koog since it is Kotlin first, with great DSLs and already have multiplatform support.

Here is an example we have in our production code, so you know what it looks like:

@Serializable
@LLMDescription("The chosen size result for an item")
data class SizeResult(
    @property:LLMDescription("ID of the size that matches the item")
    val sizeId: Long,
)

val executor = simpleOpenAIExecutor(apiKey)

override suspend fun fixWrongSize(
    sizeGroup: List<SizeGroup>,
    previousSelectedSize: Size?,
    itemDescription: String,
    userId: Long,
): SizeResult {
    val prompt = AiPrompts.sizePrompt(
        availableSizesGroup = sizeGroup,
        previousSize = previousSelectedSize?.title ?: "Unknown",
        itemDescription = itemDescription,
    )

    val responseStructure = JsonStructuredData.createJsonStructure<SizeResult>(
        schemaFormat = JsonSchemaGenerator.SchemaFormat.JsonSchema,
        examples = listOf(
            SizeResult(123)
        ),
        schemaType = JsonStructuredData.JsonSchemaType.FULL
    )

    val result = executor.executeStructured(
        prompt = prompt("fix-wrong-selected-size", LLMParams(temperature = 0.1)) {
            system(prompt)
        },
        mainModel = model,
        structure = responseStructure,
    )

    return result.getOrThrow().structure.also {
        logger.info(prettyJson.encodeToString(it))
    }
}

fun sizePrompt(
    availableSizesGroup: List<SizeGroup>,
    previousSize: String,
    itemDescription: String,
): TextContentBuilder.() -> Unit = {
    markdown {
        +"You are a clothing image analysis assistant. Your job is to extract a structured JSON object with all the metadata of the product visible in the images. In the previous step you choose a size in a bad group. Here is the list of available sizes for the category you selected."
        newline()

        +"You chose this size, which was not for the correct category: $previousSize"
        newline()

        +"Here is the description of the item you are trying to fix:"
        +itemDescription
        newline()

        +"Please choose a size in the list below, which are the only available option for the chosen category:"
        bulleted {
            availableSizesGroup.forEach {
                item(Serialization.json.encodeToString(it))
            }
        }
    }
}