Part of the Moxin Organization open-source AI ecosystem.
MLX Inference · API Server · Studio Desktop App
OminiX is an open-source ecosystem for running multimodal AI entirely on-device — LLMs, image generation, voice cloning, and speech recognition — all in Rust with zero Python dependencies at runtime.
┌──────────────────────────────────────────────────────────────┐
│ OminiX-Studio Native desktop app (Rust + Makepad) │
└──────────────────────────┬───────────────────────────────────┘
┌──────────────────────────▼───────────────────────────────────┐
│ OminiX-API OpenAI-compatible HTTP/WebSocket server │
└──────────────────────────┬───────────────────────────────────┘
┌──────────────────────────▼───────────────────────────────────┐
│ OminiX-MLX Pure-Rust inference on Apple MLX (Metal) │
└──────────────────────────────────────────────────────────────┘
| Repo | Description | Details |
|---|---|---|
| OminiX-MLX | Safe Rust bindings to Apple MLX + 14 model crates (LLM, ASR, TTS, Image Gen) | Qwen 2/3, GLM-4, Mixtral, Mistral, Paraformer, GPT-SoVITS, FLUX.2-klein, Z-Image |
| OminiX-API | OpenAI-compatible API server wrapping OminiX-MLX | Drop-in /v1/chat, /v1/audio, /v1/images, WebSocket TTS, dynamic model loading |
| OminiX-Studio | Native cross-platform desktop app built with Makepad | Chat, image gen, voice — connects to local or cloud backends |
Requirements: macOS 14+, Apple Silicon (M1–M5), Rust 1.82+, Xcode CLI Tools.
# Clone all three repos (Studio depends on API, which depends on MLX)
git clone https://github.com/OminiX-ai/OminiX-MLX.git
git clone https://github.com/OminiX-ai/OminiX-API.git
git clone https://github.com/OminiX-ai/OminiX-Studio.git
# Start the local API server (model downloads automatically)cd OminiX-API
LLM_MODEL=mlx-community/Qwen3-4B-bf16 cargo run --release &# Launch the desktop appcd ../OminiX-Studio
cargo run --releaseA native desktop app for chatting with local and cloud LLMs, generating images, and more.
git clone https://github.com/OminiX-ai/OminiX-MLX.git
git clone https://github.com/OminiX-ai/OminiX-API.git
cd OminiX-API
LLM_MODEL=mlx-community/Qwen3-4B-bf16 cargo run --releaseThis starts an OpenAI-compatible local server. Use it with any standard client:
importopenaiclient=openai.OpenAI(base_url="http://localhost:8080/v1", api_key="not-needed")
response=client.chat.completions.create(
model="qwen3",
messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)Benchmarks on Apple M3 Max (128GB):
| Task | Model | Throughput | Memory |
|---|---|---|---|
| LLM | Qwen3-4B | 45 tok/s | 8 GB |
| ASR | Paraformer | 18x real-time | 500 MB |
| TTS | GPT-SoVITS | 3-4x real-time | 2 GB |
| Image | Z-Image-Turbo | ~3s/image | 12 GB |
No Python runtime, no venvs, no GIL. Single binary via cargo build --release. Memory safety across the full stack including GPU. Cross-compilation to macOS, Windows, Linux, iOS, Android, and WebAssembly via Makepad.
Dual-licensed under MIT and Apache 2.0.
- Moxin Organization — broader AI ecosystem, open language models, and the Moly app
- Apple MLX Team — MLX framework
- Makepad — cross-platform Rust UI toolkit
- Project Robius — multi-platform Rust app framework tooling