OpenAI-Compatible Client

Lightweight client for OpenAI-compatible chat completions APIs — ideal for hybrid on-device + cloud routing.

LeapOpenAIClient / leap-openai-client (introduced in v0.10.0) is a small, dependency-light client for any OpenAI-compatible chat-completions endpoint — OpenAI itself, OpenRouter, vLLM, llama-server, or your own proxy. It ships in the same SDK release as LeapSDK, so you can route requests between an on-device LFM and a cloud model from a single app.

When to use it

• Hybrid on-device + cloud routing. Run small / fast models on-device with LeapSDK, fall back to a larger cloud model for hard prompts.
• Standardised cloud API. Talk to any OpenAI-compatible backend without pulling in a heavier OpenAI SDK.
• Streaming first. SSE streaming is the only mode — non-streaming requests aren’t exposed. streamChatCompletion(...) forces stream = true on the outgoing request regardless of the stream field on the ChatCompletionRequest you pass in.

Add the dependency

iOS / macOS (SPM)

Add the LeapOpenAIClient product to your target. See the Quick Start for the full SPM setup.

dependencies: [
    .package(url: "https://github.com/Liquid4All/leap-sdk.git", from: "0.10.8")
]

targets: [
    .target(
        name: "YourApp",
        dependencies: [
            .product(name: "LeapOpenAIClient", package: "leap-sdk"),
        ]
    )
]

In Swift sources, import LeapOpenAIClient. The Darwin (URLSession) Ktor engine is bundled — no extra HTTP setup needed.

Android (Gradle)

dependencies {
  implementation("ai.liquid.leap:leap-sdk:0.10.8")
  implementation("ai.liquid.leap:leap-openai-client:0.10.8")
}

Bundles an OkHttp-engine Ktor client. No extra HTTP setup needed.

JVM (Gradle)

dependencies {
    implementation("ai.liquid.leap:leap-sdk:0.10.8")
    implementation("ai.liquid.leap:leap-openai-client:0.10.8")
}

JVM support landed in v0.10.7 (the jvm slice was absent in the v0.10.0–v0.10.6 cascade). Pure-Maven JVM projects should consume the -jvm classifier directly: ai.liquid.leap:leap-openai-client-jvm:0.10.8. Bundles the CIO Ktor engine.

Kotlin/Native (Gradle)

dependencies {
    implementation("ai.liquid.leap:leap-sdk:0.10.8")
    implementation("ai.liquid.leap:leap-openai-client:0.10.8")
}

Targets linuxX64, linuxArm64, mingwX64 (Windows native), and wasmJs (browser via Ktor Js engine, added in v0.10.7).

Basic usage

Swift (iOS / macOS)

New in v0.10.8. SKIE is now applied to leap-sdk-openai-client, matching LeapSDK / LeapModelDownloader / LeapUI. Swift consumers get a real AsyncSequence, exhaustive onEnum(of:) switching, nested Kotlin class names (ChatCompletionEvent.Delta instead of the previously flattened ChatCompletionEventDelta), and an OpenAiClient(config:) convenience init — no more OpenAiClientKt. prefix. If you need the pre-SKIE manual-collector surface frozen for some reason, pin to 0.10.7; otherwise use the v0.10.8 surface below.

import LeapOpenAIClient

let client = OpenAiClient(
    config: OpenAiClientConfig(
        apiKey: "sk-…",
        baseUrl: "https://api.openai.com/v1"
    )
)

let request = ChatCompletionRequest(
    model: "gpt-4o-mini",
    messages: [
        ChatMessage.System(content: "You are a helpful assistant."),
        ChatMessage.User(content: "What is the capital of Japan?")
    ],
    temperature: 0.7
)

for try await event in client.streamChatCompletion(request: request) {
    switch onEnum(of: event) {
    case .delta(let delta):
        print(delta.content, terminator: "")
    case .done(let done):
        if let usage = done.usage { print("\nTokens: \(usage.totalTokens)") }
    case .error(let err):
        print("\nError: \(err.message)")
    }
}

client.close()  // closes the underlying URLSession-backed HttpClient

onEnum(of:) gives exhaustive switching — the Swift compiler errors if a new ChatCompletionEvent case is added.

SKIE bridges Kotlin Flow to a Swift AsyncSequence with Failure = Never, so transport-level failures (network drop, TLS error, etc.) silently terminate the stream rather than throwing. Only HTTP-level errors (non-2xx response) arrive as in-band .error events; malformed SSE chunks are logged and skipped. If you need to react to silent termination, track whether .done was observed and treat a stream that ended without it as a transport failure.

Kotlin (all platforms)

import ai.liquid.leap.openai.ChatCompletionEvent
import ai.liquid.leap.openai.ChatCompletionRequest
import ai.liquid.leap.openai.ChatMessage
import ai.liquid.leap.openai.OpenAiClient
import ai.liquid.leap.openai.OpenAiClientConfig

val client = OpenAiClient(
    config = OpenAiClientConfig(
        apiKey = "sk-…",
        baseUrl = "https://api.openai.com/v1",
    )
)

val request = ChatCompletionRequest(
    model = "gpt-4o-mini",
    messages = listOf(
        ChatMessage.System("You are a helpful assistant."),
        ChatMessage.User("What is the capital of Japan?"),
    ),
    temperature = 0.7,
)

client.streamChatCompletion(request).collect { event ->
    when (event) {
        is ChatCompletionEvent.Delta -> print(event.content)
        is ChatCompletionEvent.Done  -> event.usage?.let { println("\nTokens: ${it.totalTokens}") }
        is ChatCompletionEvent.Error -> println("\nError: ${event.message}")
    }
}

client.close()

Configuration

OpenAiClientConfig is a Kotlin data class bridged identically on every platform.

data class OpenAiClientConfig(
    val apiKey: String,
    val baseUrl: String = "https://api.openai.com/v1",
    val chatCompletionsPath: String = "/chat/completions",
    val extraHeaders: Map<String, String> = emptyMap(),
)

Field	Default	Notes
apiKey	— (required)	Sent as Authorization: Bearer <apiKey>.
baseUrl	https://api.openai.com/v1	Override for OpenRouter, a self-hosted backend, etc.
chatCompletionsPath	/chat/completions	Appended to baseUrl.
extraHeaders	{}	Merged into every request — e.g. OpenRouter’s HTTP-Referer.

OpenRouter

Swift (iOS / macOS)

let client = OpenAiClient(
    config: OpenAiClientConfig(
        apiKey: "sk-or-…",
        baseUrl: "https://openrouter.ai/api/v1",
        extraHeaders: [
            "HTTP-Referer": "https://yourapp.example.com",
            "X-Title": "Your App"
        ]
    )
)

Kotlin (all platforms)

val client = OpenAiClient(
    OpenAiClientConfig(
        apiKey = "sk-or-…",
        baseUrl = "https://openrouter.ai/api/v1",
        extraHeaders = mapOf(
            "HTTP-Referer" to "https://yourapp.example.com",
            "X-Title" to "Your App",
        ),
    )
)

Self-hosted vLLM / llama-server

Swift (iOS / macOS)

let client = OpenAiClient(
    config: OpenAiClientConfig(
        apiKey: "anything",  // Required by config but typically unused
        baseUrl: "http://10.0.0.42:8000/v1"
    )
)

Kotlin (all platforms)

val client = OpenAiClient(
    OpenAiClientConfig(
        apiKey = "anything",
        baseUrl = "http://10.0.0.42:8000/v1",
    )
)

Request shape

ChatCompletionRequest covers standard OpenAI fields plus a few OpenRouter-specific extensions. OpenRouter-only fields are silently ignored by stock OpenAI-compatible APIs.

data class ChatCompletionRequest(
    val model: String,
    val messages: List<ChatMessage>,
    val temperature: Double? = null,
    val topP: Double? = null,
    val maxCompletionTokens: Int? = null,   // Preferred for newer OpenAI versions
    val maxTokens: Int? = null,             // Legacy alias — some custom backends still require it
    val frequencyPenalty: Double? = null,
    val presencePenalty: Double? = null,
    val stop: List<String>? = null,
    val stream: Boolean = true,
    // OpenRouter extensions:
    val topK: Int? = null,
    val repetitionPenalty: Double? = null,
    val minP: Double? = null,
    val topA: Double? = null,
    val transforms: List<String>? = null,
    val models: List<String>? = null,
    val route: String? = null,
    val provider: ProviderPreferences? = null,
)

ChatMessage (the OpenAI-client one, distinct from LeapSDK.ChatMessage) is a sealed type with three cases — System, User, Assistant.

Response shape

streamChatCompletion(request) returns a Flow<ChatCompletionEvent> (Kotlin) — SKIE bridges this as a Swift AsyncSequence since v0.10.8, so Swift consumers can iterate it with for try await event in client.streamChatCompletion(request: ...). Events:

Variant	Meaning
Delta(content: String)	Text chunk from the model. May be empty for role-only deltas.
Done(usage: Usage?)	Stream finished. usage is non-null when the API includes token counts.
Error(message: String)	Non-2xx HTTP response. Malformed SSE chunks are logged and skipped — they do not emit an Error event.

data class Usage(val promptTokens: Int, val completionTokens: Int, val totalTokens: Int)

Hybrid routing example

Route simple prompts to a small on-device LFM; escalate harder prompts to a cloud model.

Swift (iOS / macOS)

import LeapModelDownloader
import LeapOpenAIClient

@MainActor
final class HybridChatViewModel: ObservableObject {
    private let onDevice: Conversation
    private let cloud: OpenAiClient

    init(onDevice: Conversation, cloud: OpenAiClient) {
        self.onDevice = onDevice
        self.cloud = cloud
    }

    func send(_ text: String, useCloud: Bool) async throws {
        if useCloud {
            // Cloud path — same SKIE-bridged surface as on-device since v0.10.8.
            let request = ChatCompletionRequest(
                model: "gpt-4o-mini",
                messages: [ChatMessage.User(content: text)]
            )
            for try await event in cloud.streamChatCompletion(request: request) {
                if case let .delta(d) = onEnum(of: event) { appendChunk(d.content) }
            }
        } else {
            // On-device path.
            let userMessage = LeapSDK.ChatMessage(role: .user, textContent: text)
            for try await response in onDevice.generateResponse(message: userMessage) {
                if case let .chunk(c) = onEnum(of: response) { appendChunk(c.text) }
            }
        }
    }

    private func appendChunk(_ text: String) { /* … */ }

    deinit { cloud.close() }
}

Kotlin (Android)

import ai.liquid.leap.Conversation
import ai.liquid.leap.message.MessageResponse
import ai.liquid.leap.openai.ChatCompletionEvent
import ai.liquid.leap.openai.ChatCompletionRequest
import ai.liquid.leap.openai.ChatMessage as CloudChatMessage
import ai.liquid.leap.openai.OpenAiClient
import ai.liquid.leap.message.ChatMessage
import ai.liquid.leap.message.ChatMessageContent
import androidx.lifecycle.ViewModel
import androidx.lifecycle.viewModelScope
import kotlinx.coroutines.launch

class HybridChatViewModel(
    private val onDevice: Conversation,
    private val cloud: OpenAiClient,
) : ViewModel() {

    fun send(text: String, useCloud: Boolean) = viewModelScope.launch {
        if (useCloud) {
            val request = ChatCompletionRequest(
                model = "gpt-4o-mini",
                messages = listOf(CloudChatMessage.User(text)),
            )
            cloud.streamChatCompletion(request).collect { event ->
                if (event is ChatCompletionEvent.Delta) appendChunk(event.content)
            }
        } else {
            val message = ChatMessage(
                role = ChatMessage.Role.USER,
                content = listOf(ChatMessageContent.Text(text)),
            )
            onDevice.generateResponse(message).collect { resp ->
                if (resp is MessageResponse.Chunk) appendChunk(resp.text)
            }
        }
    }

    private fun appendChunk(text: String) { /* … */ }

    override fun onCleared() {
        super.onCleared()
        cloud.close()
    }
}

Kotlin (JVM / native)

suspend fun hybridSend(
    onDevice: Conversation,
    cloud: OpenAiClient,
    text: String,
    useCloud: Boolean,
) {
    if (useCloud) {
        val request = ChatCompletionRequest(
            model = "gpt-4o-mini",
            messages = listOf(CloudChatMessage.User(text)),
        )
        cloud.streamChatCompletion(request).collect { event ->
            if (event is ChatCompletionEvent.Delta) print(event.content)
        }
    } else {
        onDevice.generateResponse(text).collect { resp ->
            if (resp is MessageResponse.Chunk) print(resp.text)
        }
    }
}

See Cloud AI Comparison for a side-by-side feature breakdown.

Lifecycle

The OpenAiClient(config:) factory (Kotlin fun OpenAiClient(config:) — exported as a SKIE-bundled Swift convenience init since v0.10.8) creates an HttpClient internally and ties it to the returned client — call close() when you’re done.

Swift (iOS / macOS)

deinit { client.close() }

The lower-level constructor that accepts an externally-managed HttpClient is part of the Kotlin/Ktor surface and isn’t a useful entry point from Swift — the Ktor engine machinery isn’t bridged into the public Swift API. Use OpenAiClient(config:) and let the SDK own the session. If multiple consumers share a client, share the OpenAiClient instance and close() once at teardown.

Kotlin (all platforms)

override fun onCleared() {
    super.onCleared()
    client.close()
}

If you need to share an HttpClient across multiple clients (e.g., you already manage one for other Ktor-based code), use the lower-level constructor that takes a pre-built HttpClient — you then own its lifetime and shouldn’t call close() on the OpenAiClient:

val shared = HttpClient(OkHttp)  // your own instance
val client = OpenAiClient(config = config, httpClient = shared)
// Don't call client.close() — you own `shared` and decide when it dies