Getting started on Llamatik Documentation

Installation

Mon, 01 Jan 0001 00:00:00 +0000

**⚠️ Legacy (v1.x)** — You are viewing the old documentation. [View current v2.0 docs](/)

Gradle dependency#

Add the library to the module where you want to use Llamatik.

dependencies {
 implementation("com.llamatik:library:0.x.y")
}

Kotlin Multiplatform#

In a KMP project, add the dependency to the source sets that need it.

kotlin {
 sourceSets {
 val commonMain by getting {
 dependencies {
 implementation("com.llamatik:library:0.x.y")
 }
 }
 }
}

Platform notes#

Android#

Min SDK: 26
The library loads native code at runtime.
If you package models in assets, make sure your app can resolve or copy them to a readable file path before initialization.

iOS#

Minimum deployment target: 16.6
The library is consumed through Kotlin Multiplatform / Kotlin Native integration.
Model files need to be bundled or downloaded in a way your app can resolve to an actual file path.

JVM / Desktop#

Use a recent JDK. The repository is configured around toolchain 21.
Native libraries are loaded from packaged resources.
Model paths are usually absolute disk paths.

WASM#

Text generation is available through the WASM implementation.
Embeddings, Stable Diffusion, Whisper, and KV session persistence are currently not available.
If you rely on streaming or worker-only execution, prefer the streaming APIs.

What to install beyond the dependency#

Llamatik gives you the native bridge. You still need compatible model files:

Quickstart

Mon, 01 Jan 0001 00:00:00 +0000

**⚠️ Legacy (v1.x)** — You are viewing the old documentation. [View current v2.0 docs](/)

This page shows the most common first steps with the library.

1. Resolve a model path#

import com.llamatik.library.platform.LlamaBridge

val modelPath = LlamaBridge.getModelPath("tinyllama.gguf")

On desktop this is often just a file path. On mobile you may resolve the file from app storage, assets, or a download location.

2. Load a generation model#

val ok = LlamaBridge.initGenerateModel(modelPath)
check(ok) { "Failed to init model at $modelPath" }

3. Generate text#

val text = LlamaBridge.generate("Write a 2 sentence story about a llama.")
println(text)

4. Stream text#

LlamaBridge.generateStream(
 prompt = "Stream a short haiku.",
 callback = object : GenStream {
 override fun onDelta(text: String) = print(text)
 override fun onComplete() = println("\nDone")
 override fun onError(message: String) = println("Error: $message")
 }
)

5. Use embeddings#

val embedModelPath = LlamaBridge.getModelPath("nomic-embed-text.gguf")
val embedOk = LlamaBridge.initEmbedModel(embedModelPath)
check(embedOk) { "Failed to init embedding model" }

val vector: FloatArray = LlamaBridge.embed("hello embeddings")
println("dims = ${vector.size}")

6. Continue a session#

LlamaBridge.initGenerateModel(modelPath)

val answer1 = LlamaBridge.generate("Explain Kotlin coroutines.")
val answer2 = LlamaBridge.generateContinue("Now show a very short example.")

7. Save and restore a session#

val sessionPath = "/tmp/chat.session"

LlamaBridge.sessionSave(sessionPath)
LlamaBridge.sessionLoad(sessionPath)

8. Generate an image#

import com.llamatik.library.platform.StableDiffusionBridge

val sdPath = StableDiffusionBridge.getModelPath("dreamshaper.safetensors")
check(StableDiffusionBridge.initModel(sdPath))

val rgba = StableDiffusionBridge.txt2img(
 prompt = "A watercolor painting of a mountain village"
)
println("Generated bytes = ${rgba.size}")

9. Transcribe audio#

import com.llamatik.library.platform.WhisperBridge

val whisperPath = WhisperBridge.getModelPath("ggml-base.en.bin")
check(WhisperBridge.initModel(whisperPath))

val textResult = WhisperBridge.transcribeWav("/path/to/audio.wav", language = "en")
println(textResult)

From here, move on to the Guides section for more detailed patterns.

Models

Mon, 01 Jan 0001 00:00:00 +0000

**⚠️ Legacy (v1.x)** — You are viewing the old documentation. [View current v2.0 docs](/)

Model selection has the biggest impact on speed, memory usage, and output quality.

LlamaBridge models#

For text generation and embeddings, Llamatik works with GGUF models.

Text generation#

Choose an instruction-tuned GGUF model when building chat, assistants, extraction, or summarization features.

Embeddings#

Use a model specifically intended for embeddings when calling initEmbedModel(...) and embed(...). Do not assume your generation model is a good embedding model.