Chat

There is no chat() function. Single-turn generation is RunAnywhere.generate(prompt, options). Multi-turn conversations pass prior messages through LLMGenerateRequest.history. All LLM entry points are suspend extension functions on RunAnywhere, imported from com.runanywhere.sdk.public.extensions.

Single turn

import com.runanywhere.sdk.public.RunAnywhere
import com.runanywhere.sdk.public.extensions.generate

val result = RunAnywhere.generate("What is the capital of France?")
println(result.text)

generate returns LLMGenerationResult. Wire-generated proto fields are snake_case: result.text, result.tokens_per_second, result.input_tokens, result.tokens_generated, result.generation_time_ms.

Multi-turn

import ai.runanywhere.proto.v1.ChatMessage
import ai.runanywhere.proto.v1.LLMGenerateRequest
import ai.runanywhere.proto.v1.MessageRole

val history = listOf(
    ChatMessage(role = MessageRole.MESSAGE_ROLE_USER, content = "My name is Ada."),
    ChatMessage(role = MessageRole.MESSAGE_ROLE_ASSISTANT, content = "Nice to meet you, Ada."),
)

val result = RunAnywhere.generate(
    LLMGenerateRequest(prompt = "What is my name?", history = history)
)
println(result.text)

MessageRole values: MESSAGE_ROLE_UNSPECIFIED, MESSAGE_ROLE_USER, MESSAGE_ROLE_ASSISTANT, MESSAGE_ROLE_SYSTEM, MESSAGE_ROLE_TOOL, MESSAGE_ROLE_DEVELOPER. Wire generates plain classes with all-defaulted constructor parameters plus copy(). Use named arguments; there is no builder.

System prompt

import ai.runanywhere.proto.v1.LLMGenerationOptions

val options = LLMGenerationOptions(
    system_prompt = "You are a terse assistant. Answer in one sentence.",
    max_tokens = 256,
    temperature = 0.7f,
)

val result = RunAnywhere.generate("Explain HTTP caching.", options)

Keep the persona on the options rather than in history so history trimming cannot drop it.

Conversation cache

val request = LLMGenerateRequest(
    prompt = userInput,
    history = history,
    conversation_id = sessionId,
)

A stable conversation_id lets backends that keep a prompt cache reuse it across turns.

Streaming a turn

generateStream returns a Flow<LLMStreamEvent> and is not a suspend function.

import com.runanywhere.sdk.public.extensions.generateStream

val reply = StringBuilder()
RunAnywhere.generateStream(userInput).collect { event ->
    reply.append(event.token)
}

Trimming history

val maxTurns = 10
val trimmed = history.takeLast(maxTurns * 2)

Getting Started

Swift SDK

Kotlin SDK

React Native SDK

Flutter SDK

Web SDK

Vibe Coding

Single turn

Multi-turn

System prompt

Conversation cache

Streaming a turn

Trimming history

​Single turn

​Multi-turn

​System prompt

​Conversation cache

​Streaming a turn

​Trimming history

Single turn

Multi-turn

System prompt

Conversation cache

Streaming a turn

Trimming history