xybrid library

Xybrid - Hybrid cloud-edge ML inference orchestrator.

This library provides a clean Dart API for running ML models on-device or in the cloud, with intelligent routing based on device capabilities.

Initialization

You must call Xybrid.init once before using any Xybrid functionality:

void main() async {
  await Xybrid.init();
  runApp(MyApp());
}

Quick Start

// Initialize once at app startup
await Xybrid.init();

// Load a model from registry
final loader = XybridModelLoader.fromRegistry('kokoro-82m');
final model = await loader.load();

// Run text-to-speech
final envelope = XybridEnvelope.text('Hello, world!');
final result = await model.run(envelope);
final audioBytes = result.audioBytes;

Pipelines

// Load and run a pipeline
final pipeline = XybridPipeline.fromYaml('''
name: speech-to-text
stages:
  - model: whisper-small
''');
final envelope = XybridEnvelope.audio(bytes: audioData, sampleRate: 16000);
final result = await pipeline.run(envelope);

Conversation Context (LLM)

// Create conversation context for multi-turn chat
final context = ConversationContext();
context.setSystem('You are a helpful assistant.');

// Run with context
context.pushText('Hello!', MessageRole.user);
final result = await model.runWithContext(
  XybridEnvelope.text('Hello!'),
  context,
);
context.pushText(result.text ?? '', MessageRole.assistant);

Classes

• XybridModelLoader - Prepare and load models from registry or bundles
• XybridModel - A loaded model ready for inference
• XybridEnvelope - Input data wrapper for different modalities
• XybridResult - Inference output containing text, audio, or embeddings
• XybridPipeline - Multi-stage inference pipeline
• ConversationContext - Multi-turn conversation memory for LLMs
• MessageRole - Message role for conversation turns
• GenerationConfig - LLM generation parameters (temperature, top-p, etc.)

Classes

ConversationContext

Manages multi-turn conversation history for LLM models.

GenerationConfig

Generation parameters for LLM inference.

LoadComplete

Model loading completed successfully.

LoadError

Model loading failed with an error.

LoadEvent

Event emitted during model loading with progress tracking.

LoadProgress

Download progress update (0.0 to 1.0).

StreamToken

A single token emitted during streaming generation.

Xybrid

Main entry point for the Xybrid SDK.

XybridEnvelope

Envelope containing input data for model inference.

XybridModel

A loaded model ready for inference.

XybridModelLoader

Prepares a model for loading from registry or local bundle.

XybridPipeline

A loaded pipeline ready for multi-stage inference.

XybridResult

Result of a model inference operation.

Enums

MessageRole

Message role for conversation turns.

Functions

wrapInWavHeader(Uint8List pcmBytes, {required int sampleRate, required int channels}) → Uint8List

Wraps raw PCM audio bytes in a WAV header for playback.

Exceptions / Errors

XybridException

Exception thrown when Xybrid operations fail.