synthetic

A Kotlin-first LLM toolkit offering unified API for AI models (OpenAI, HuggingFace, DeepSeek, custom) and modular data storage. Features agent-enabled architecture with async RAG capabilities and zero boilerplate.

查看中文介绍

Synthetic is a lightweight Kotlin/JVM library for building LLM-driven applications, specializing in RAG ( Retrieval-Augmented Generation) pipelines and agent systems. Designed for API flexibility, it implements an adapter-first architecture to seamlessly integrate any AI provider (OpenAI, HuggingFace, custom endpoints) and data storage system.

Core features:

• Modular RAG components: Document processors, embedding interfaces, and retriever/reranker blueprints
• Flexible Agent framework: Custom tool integration, stateful conversation management, and goal-oriented execution flows
• Agnostic data layer: Abstracted connectors for vector DBs, SQL/NoSQL, or file systems via unified interfaces
• API-neutral LLM operations: Swap models/endpoints without rewriting chains/agents
• Coroutine-native pipelines: Async workflows for both RAG and Agent operations
• Minimal abstraction overhead: Clear extension points for custom agent logic and data handlers

Optimized for Kotlin/JVM developers prioritizing API interoperability and custom data flows over framework rigidity.

Usage

Common

private val apiKey = "sk-xxxxxxxxxx"
private val url = Url("https://synthetic.sithra.org/v1")
fun main() = runBlocking {
    val adapter = OpenAIAdapter {
        baseUrl = url
        token = apiKey
    }
    val ctx = Context {
        apiAdapter = adapter
        messagesStore = MessagesStoreInMemory()
    }
    val session = ctx.createSession("demo-model")
    // common
    ctx.withSession(session) {
        val message = chat(Message("why hello world?"))
        println(message.content)
        addMessage(message)
    }

    // streaming
    ctx.withSession(session) {
        val response = chatStream(Message("why hello world?"))