Physical benchmark
PhyGround
A benchmark for the physical plausibility of text-and-image-to-video generation, with 250 curated prompts, a 13-law physics taxonomy, and the open PhyJudge-9B evaluator.

AI Crafted in Pure Rust
Native Desktop Intelligence — No Python Required
Like the ancient seal carvers who shaped meaning into stone with precision and intent, we craft AI tools with the same deliberate mastery.
Every line of code, from UI to inference, written in Rust. No Python runtime, no garbage collection pauses, no compromise.
Direct GPU access through Apple Metal. Models run at native speed on Apple Silicon, with zero abstraction overhead.
Your data never leaves your machine. Full AI capabilities running natively on your desktop, private by design.
Native AI applications built with the Moxin stack — each one a testament to what's possible when performance meets purpose.
A native desktop application that runs LLMs, generates images, clones voices, and transcribes speech — all locally on Apple Silicon. No cloud, no Python, no waiting.
▶ Overview playing — click a feature to explore
View on GitHubLive translation with real-time bilingual subtitles at its core. Clone any voice from seconds of audio and generate natural speech across Chinese, English, Japanese, and Korean.
▶ Overview playing — click a feature to explore
View on GitHubMoxin connects open AI applications with research into systems that simulate, predict, and act in the physical world. Explore the latest releases from NU World Model & Embodied AI.
Physical benchmark
A benchmark for the physical plausibility of text-and-image-to-video generation, with 250 curated prompts, a 13-law physics taxonomy, and the open PhyJudge-9B evaluator.
Video world model
A physics-aware video-generation world model post-trained with temporal-coherence tuning and preference optimization derived from the PhyGround human-annotation pool.
A shared research loop
PhyGround provides the physical benchmark and human preference data; PhyWorld uses that signal to build a more physically coherent video-generation world model.
A vertically integrated stack from metal to pixel. Every component chosen for performance, every abstraction earned.
Systems language powering every layer
Memory safety without garbage collection
GPU-accelerated native UI framework
Pure Rust, cross-platform rendering
Apple Silicon inference engine
Metal GPU acceleration for ML models
Dataflow orchestration framework
Connecting AI pipelines seamlessly
Every application in the Moxin ecosystem is open source under Apache 2.0. Inspect the code, contribute improvements, or build your own tools on our stack.