A native desktop AI app built with pure Rust and Makepad.

Run local LLMs, generate images, clone voices, and transcribe speech — all on your own hardware, without a Python runtime.

Getting Started | Supported Models | Platform

---

Getting Started

Requirements

• macOS 14.0+ (Sonoma) on Apple Silicon (M1-M5)
• Rust 1.82+
• Xcode Command Line Tools (xcode-select --install)

1. Install OminiX-API

Moxin Studio uses OminiX-API as the local inference server. Install it first:

curl -fsSL https://raw.githubusercontent.com/OminiX-ai/OminiX-API/main/install.sh | sh

This installs ominix-api to /usr/local/bin and creates ~/.OminiX/ for config and models.

2. Build and run Moxin Studio

git clone https://github.com/moxin-org/Moxin-Studio.git
cd Moxin-Studio
cargo run -p moly-shell --bin moxin-studio

The first build takes a few minutes to compile all dependencies. Subsequent runs are fast.

3. Download a model and start chatting

Open the Model Hub from the sidebar, click Download on any model, then click Load. Moxin Studio will auto-start OminiX-API and route your chat through it.

Features

• Local AI inference — Run LLMs, vision models, image generation, speech recognition, and TTS directly on your Mac via OminiX-API
• Model Hub — Discover, download, and run models directly from the app
• Voice I/O — Speech-to-text and text-to-speech with voice cloning
• MCP support — Model Context Protocol for tool use
• Chat history — Persistent, searchable conversation history

Supported Local Models

Every model below has a dedicated, optimized implementation — not a generic wrapper. The pure Rust models run directly via OminiX-MLX with Metal GPU acceleration.

LLM — Large Language Models

Model	Implementation	Notes
Qwen3	Pure Rust	0.6B, 4B, 8B variants
Qwen3.5-27B	Pure Rust	Hybrid DeltaNet + Attention
GLM-4	Pure Rust
GLM-4.7-Flash	Pure Rust	MoE + MLA architecture
GLM-4.5 MoE	Pure Rust	Mixture of Experts
Mistral / Nemo	Pure Rust
Mixtral	Pure Rust	MoE
MiniCPM-SALA	Pure Rust	Hybrid attention

VLM — Vision Language Models

Model	Implementation
Qwen3-VL	Pure Rust
Moxin-7B	Pure Rust
DeepSeek-OCR-2	Pure Rust

ASR — Speech Recognition

Model	Implementation	Notes
Qwen3-ASR	Pure Rust	30+ languages
Paraformer (FunASR)	Pure Rust
FunASR-Nano	Pure Rust	Lightweight
SenseVoice + Qwen3-4B	Pure Rust	LLM-enhanced ASR

TTS — Text to Speech

Model	Implementation	Notes
Qwen3-TTS	Pure Rust	Preset voices + voice cloning
GPT-SoVITS	Pure Rust	Zero-shot voice cloning
Step-Audio 2	Pure Rust

Image Generation

Model	Implementation	Notes
FLUX.2-klein	Pure Rust	Also available as GGUF
Z-Image-Turbo	Pure Rust
Qwen-Image-2512	Pure Rust
Qwen-Image-Edit-2511	Python MLX	Image editing
Cosmos Predict2 14B	Python MLX	Text-to-image

Video Generation

Model	Implementation	Notes
Wan2.2 5B	Python MLX	Text-to-video

The Moxin / OminiX Platform

Moxin Studio is the user-facing layer of a three-part pure Rust AI platform:

┌─────────────────────────────────────────────┐
│            Moxin Studio (this repo)         │  Desktop UI (Rust + Makepad)
│         Chat · Models · Voice · Settings    │
└──────────────────────┬──────────────────────┘
                       │ OpenAI-compatible REST/WS
┌──────────────────────▼──────────────────────┐
│               OminiX-API                    │  Local inference server (pure Rust)
│    LLM · ASR · TTS · Image endpoints       │
└──────────────────────┬──────────────────────┘
                       │ Rust crate interface
┌──────────────────────▼──────────────────────┐
│               OminiX-MLX                    │  On-device inference backend
│      Metal-accelerated · MLX framework      │  (Apple Silicon)
└─────────────────────────────────────────────┘

• OminiX-MLX — Apple Silicon inference engine. Pure-Rust bindings to Apple's MLX framework with Metal GPU acceleration. Supports LLMs, VLMs, ASR, TTS, and image generation.
• OminiX-API — Local inference server. OpenAI-compatible HTTP and WebSocket endpoints with dynamic model loading at runtime.
• Moxin Studio (this repo) — Desktop application. Connects to OminiX-API for local inference and cloud providers for remote models.

Project Structure

Moxin-Studio/
├── moly-shell/          # Main application binary
├── moly-data/           # Shared state, persistence, API clients
├── moly-widgets/        # Reusable UI components and theming
└── apps/
    ├── moly-chat/       # Chat interface
    ├── moly-hub/        # Model Hub (discovery, download, load/unload)
    ├── moly-settings/   # Provider and API key configuration
    ├── moly-mcp/        # MCP server configuration
    └── moly-voice/      # Voice I/O

License

Apache 2.0