English · 简体中文

Xybrid

On-device AI for mobile, desktop, edge.
Run speech, language, and vision models locally — private, offline, fast.
Perfect for any app including games 🕹️

Documentation · SDKs · Models · Join Discord · Issues

---

    

SDKs

Xybrid is a Rust-powered runtime with native bindings for every major platform. Pick your SDK:

SDK	Platforms	Install	Status	Sample
Flutter	iOS, Android, macOS, Linux, Windows	pub.dev	Available	README
Unity	macOS, Windows, Linux	See below	Available	Unity 3D AI tavern
Swift	iOS, macOS	Swift Package Manager	Coming Soon	README
Kotlin	Android	Maven Central	Available	README
CLI	macOS, Linux, Windows	Download binary	Available	—
Rust	All	xybrid-core / xybrid-sdk	Available	—

Every SDK wraps the same Rust core — identical model support and behavior across all platforms.

Install

Unity — Package Manager → Add from git URL:

https://github.com/xybrid-ai/xybrid.git?path=bindings/unity

Flutter — add to your pubspec.yaml:

dependencies:
  xybrid_flutter: ^0.1.0

Kotlin (Android) — add to your build.gradle.kts:

dependencies {
    implementation("ai.xybrid:xybrid-kotlin:0.1.0-alpha7")
}

---

Quick Start

See each SDK's README for platform-specific setup: Flutter · Unity · Swift · Kotlin · Rust

Single Model

Run a model in one line from the CLI, or three lines from any SDK:

CLI:

xybrid run kokoro-82m --input "Hello world" -o output.wav

Flutter:

final model = await Xybrid.model(modelId: 'kokoro-82m').load();
final result = await model.run(envelope: Envelope.text(text: 'Hello world'));
// result → 24kHz WAV audio

Kotlin:

val model = Xybrid.model(modelId = "kokoro-82m").load()
val result = model.run(envelope = XybridEnvelope.Text("Hello world"))
// result → 24kHz WAV audio

Swift:

let model = try Xybrid.model(modelId: "kokoro-82m").load()
let result = try model.run(envelope: .text("Hello world"))
// result → 24kHz WAV audio

Unity (C#):

var model = Xybrid.Model(modelId: "kokoro-82m").Load();
var result = model.Run(envelope: Envelope.Text("Hello world"));
// result → 24kHz WAV audio

Rust:

let model = Xybrid::model("kokoro-82m").load()?;
let result = model.run(&Envelope::text("Hello world"))?;
// result → 24kHz WAV audio

Pipelines

Chain models together — build a voice assistant in 3 lines of YAML:

# voice-assistant.yaml
name: voice-assistant
stages:
  - model: whisper-tiny    # Speech → text
  - model: qwen2.5-0.5b    # Process with LLM
  - model: kokoro-82m      # Text → speech

CLI:

xybrid run voice-assistant.yaml --input question.wav -o response.wav

Flutter:

final pipeline = await Xybrid.pipeline(yamlContent: yamlString).load();
await pipeline.loadModels();
final result = await pipeline.run(envelope: Envelope.audio(bytes: audioBytes));

Kotlin:

val pipeline = Xybrid.pipeline(yamlContent = yamlString).load()
pipeline.loadModels()
val result = pipeline.run(envelope = XybridEnvelope.Audio(bytes = audioBytes))

Swift:

let pipeline = try Xybrid.pipeline(yamlContent: yamlString).load()
try pipeline.loadModels()
let result = try pipeline.run(envelope: .audio(bytes: audioBytes))

Unity (C#):

var pipeline = Xybrid.Pipeline(yamlContent: yamlString).Load();
pipeline.LoadModels();
var result = pipeline.Run(envelope: Envelope.Audio(bytes: audioBytes));

Rust:

let pipeline = Xybrid::pipeline(&yaml_string).load()?;
pipeline.load_models()?;
let result = pipeline.run(&Envelope::audio(audio_bytes))?;

---

Supported Models

All models run entirely on-device. No cloud, no API keys required. Browse the full registry with xybrid models list.

Speech-to-Text

Model	Params	Format	Description
Whisper Tiny	39M	SafeTensors	Multilingual transcription (Candle runtime)
Wav2Vec2 Base	95M	ONNX	English ASR with CTC decoding

Text-to-Speech

Model	Params	Format	Description
Kokoro 82M	82M	ONNX	High-quality, 24 natural voices
KittenTTS Nano	15M	ONNX	Ultra-lightweight, 8 voices

Language Models

Model	Params	Format	Description
Gemma 3 1B	1B	GGUF Q4_K_M	Google's mobile-optimized LLM
Llama 3.2 1B	1B	GGUF Q4_K_M	Meta's general purpose, 128K context
Qwen 2.5 0.5B	500M	GGUF Q4_K_M	Compact on-device chat
SmolLM2 360M	360M	GGUF Q4_K_M	Best tiny LLM, excellent quality/size ratio

Coming Soon

Model	Type	Params	Priority	Status
Phi-4 Mini	LLM	3.8B	P2	Spec Ready (first multi-quant: Q4, Q8, FP16)
Qwen3 0.6B	LLM	600M	P2	Planned
Trinity Nano	LLM (MoE)	6B (1B active)	P2	Planned
LFM2 700M	LLM	700M	P2	Planned
Nomic Embed Text v1.5	Embeddings	137M	P1	Blocked (needs Tokenize/MeanPool steps)
LFM2-VL 450M	Vision	450M	P2	Planned
Whisper Tiny CoreML	ASR	39M	P2	Planned
Qwen3-TTS 0.6B	TTS	600M	P2	Blocked (needs custom SafeTensors runtime)
Chatterbox Turbo	TTS	350M	P3	Blocked (needs ModelGraph template)

---

Features

Capability	iOS	Android	macOS	Linux	Windows
Speech-to-Text	✅	✅	✅	✅	✅
Text-to-Speech	✅	✅	✅	✅	✅
Language Models	✅	✅	✅	✅	✅
Vision Models	✅	✅	✅	✅	✅
Embeddings	✅	✅	✅	✅	✅
Pipeline Orchestration	✅	✅	✅	✅	✅
Model Download & Caching	✅	✅	✅	✅	✅
Hardware Acceleration	Metal, ANE	CPU	Metal, ANE	CUDA	CUDA

---

Why Xybrid?

• Privacy first — All inference runs on-device. Your data never leaves the device.
• Offline capable — No internet required after initial model download.
• Cross-platform — One API across iOS, Android, macOS, Linux, and Windows.
• Pipeline orchestration — Chain models together (ASR → LLM → TTS) in a single call.
• Automatic optimization — Hardware acceleration on Apple Neural Engine, Metal, and CUDA.

---

Community

• Documentation
• Discord
• GitHub Issues

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines on setting up your development environment, submitting pull requests, and adding new models.

Star History

License

Apache License 2.0 — see LICENSE for details.