API reference on Llamatik Documentation

LlamaBridge

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

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

LlamaBridge is the main multiplatform entry point for GGUF-based language models in Llamatik. It exposes a compact API for:

text generation
retrieval-friendly embeddings
streaming tokens
JSON and JSON Schema constrained output
generation parameter tuning
KV cache reuse and session persistence
model introspection (finetune type, chat template)
concurrent independent sessions via LlamaSession
Multi-Token Prediction (MTP) for faster speculative generation

Under the hood, the actual implementation is platform-specific, but the public API stays the same.

GenStream

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

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

GenStream is the callback interface used by Llamatik’s streaming generation APIs.

interface GenStream {
 fun onDelta(text: String)
 fun onComplete()
 fun onError(message: String)
}

When to use it#

Use GenStream when you want to show output progressively instead of waiting for the full answer. This is the typical choice for chat UIs, terminals, typing animations, or any screen where responsiveness matters.

StableDiffusionBridge

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

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

StableDiffusionBridge provides on-device text-to-image generation. It is intended for apps that want local image generation without sending prompts or generated images to external servers.

Platform support#

Current support in the library code:

Android: supported
iOS: supported
JVM / Desktop: supported
WASM: currently not implemented

Public API#

expect object StableDiffusionBridge {
 fun getModelPath(modelFileName: String): String
 fun initModel(modelPath: String, threads: Int = -1): Boolean
 fun txt2img(
 prompt: String,
 negativePrompt: String? = null,
 width: Int = 512,
 height: Int = 512,
 steps: Int = 20,
 cfgScale: Float = 7.0f,
 seed: Long = -1L,
 ): ByteArray
 fun img2img(
 initImageRgba: ByteArray,
 initImageW: Int,
 initImageH: Int,
 prompt: String,
 negativePrompt: String? = null,
 width: Int = 512,
 height: Int = 512,
 steps: Int = 20,
 cfgScale: Float = 7.0f,
 strength: Float = 0.75f,
 seed: Long = -1L,
 ): ByteArray
 fun release()
}

`getModelPath(modelFileName)`#

Returns a usable path for the model file on the current platform.

WhisperBridge

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

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

WhisperBridge provides on-device speech-to-text using Whisper models. Its API is intentionally small: initialize a model, transcribe a WAV file, and release resources when done.

Platform support#

Platform	Status
Android	supported
iOS	supported
JVM / Desktop	supported
WASM	not implemented

Public API#

expect object WhisperBridge {
 fun getModelPath(modelFileName: String): String
 fun initModel(modelPath: String): Boolean
 fun transcribeWav(
 wavPath: String,
 language: String? = null,
 initialPrompt: String? = null,
 ): String
 fun transcribeWavSegments(
 wavPath: String,
 language: String? = null,
 initialPrompt: String? = null,
 translate: Boolean = false,
 diarize: Boolean = false,
 ): String
 fun release()
}

`getModelPath(modelFileName)`#

Returns a platform-usable path for the Whisper model file.

MultimodalBridge

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

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

MultimodalBridge provides on-device vision-language model (VLM) inference. It combines a language model with a multimodal projector to analyze images and answer questions about them.

Platform support#

Android: supported
iOS: supported
JVM / Desktop: supported
WASM: model paths are stored, but analysis is dispatched to a Web Worker — the worker must be wired up separately

Public API#

expect object MultimodalBridge {
 fun initModel(modelPath: String, mmprojPath: String): Boolean
 fun analyzeImageBytesStream(
 imageBytes: ByteArray,
 prompt: String,
 callback: GenStream,
 )
 fun cancelAnalysis()
 fun release()
}

`initModel(modelPath, mmprojPath)`#

Loads both the language model and the multimodal projector (mmproj). Returns true on success.

LlamaSession

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

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

LlamaSession is an isolated inference session with its own KV cache and cancel flag. It is the right tool when you need multiple independent generation streams running concurrently.

Create sessions via LlamaBridge.createSession(). Dispose with close() when done.

Platform support#

Android: supported
iOS: supported
JVM / Desktop: supported
WASM: currently not implemented — createSession() returns null

Public API#

expect class LlamaSession {
 val name: String
 fun stream(prompt: String, callback: GenStream)
 fun cancel()
 fun close()
}

Creating a session#

Sessions are created from LlamaBridge after a generation model has been loaded. Pass a human-readable name to identify the session in your UI:

API reference on Llamatik Documentation

LlamaBridge

GenStream

When to use it#

StableDiffusionBridge

Platform support#

Public API#

getModelPath(modelFileName)#

WhisperBridge

Platform support#

Public API#

getModelPath(modelFileName)#

MultimodalBridge

Platform support#

Public API#

initModel(modelPath, mmprojPath)#

LlamaSession

Platform support#

Public API#

Creating a session#

`getModelPath(modelFileName)`#

`getModelPath(modelFileName)`#

`initModel(modelPath, mmprojPath)`#