Vincent Gourbin
I get the latest AI models running on Apple silicon, and turn them into apps that need no server.
I port models released by AI labs, image, video, speech, music, to Swift and MLX. I publish those ports as open source, then build them into apps for Mac, iPhone and iPad. Your data stays on your device.
The forge
Each row follows one model, from its release to the app that uses it.
- Released model : FLUX.2
- Released model : LTX-2
- Released model : Gemma 4
- Released model : YuE2
- Released model : Voxtral
- Released model : RMBG 2.0
- Released model : Qwen3.8Swift / MLX port : qwen38-mlx-swiftIn the app : not in an app yet
- Released model : MiniMax-H3In the app : not in an app yet
Apps
Four apps, all computed on the device.
An AI image and video studio that runs on your Mac.
- On
- macOS
- Price
- Free
- Version
- 3.2, 15 September 2026

One sentence or a photo, and PocketAnthem turns it into a song, generated on your device.
- On
- iPhone, iPad, Mac
- Price
- Free
- Status
- Coming soon to the App Store
See the app pageComing soon to the App Store

Your meetings transcribed and summarized on your device, offline.
- On
- iPhone, iPad, Mac, Apple Watch
- Price
- €4.99
- Version
- 1.2.6, 13 November 2025

A reverse music quiz: you get what a song is about, and you have to name it.
- On
- iPhone, iPad
- Price
- Free
- Version
- 1.1, 15 December 2025


Log
Collected automatically from GitHub, Hugging Face and the App Store on 12 October 2026.
- mlx-voxtral-swift
Native Swift implementation of Voxtral speech-to-text for Apple Silicon with MLX
- yue2-mlx-swift
Swift/MLX port of YuE2 (m-a-p/YuE2-3B + YuE2-Vae) for Apple Silicon: lyrics + style -> song, on Mac and iPhone. Powers PocketAnthem on the App Store.
- gemma-4-swift-mlx
Native Gemma 4 multimodal (text+vision+audio+video) inference for Apple Silicon via MLX Swift
- qwen38-mlx-swift
Native Qwen3.8 inference on Apple Silicon with MLX Swift — dense 27B, Bonsai 2 and Flash-Next MoE; embeddable Qwen38Brain library (fast/lean profiles, conversation reuse), OpenAI-compatible server with tool calling, LangWatch-benchmarked agent loop
- swift-mlx-profiler
MLX performance profiling for Apple Silicon — GPU/CPU utilization, Chrome Trace (Perfetto), benchmarks
- Fluxforge Studio 3.2
New version on the App Store
- flux-2-swift-mlx
Flux.2 image generation on Mac with MLX - Native Swift implementation
Open source
The most followed ports on GitHub.
- flux-2-swift-mlx
Flux.2 image generation on Mac with MLX - Native Swift implementation
- gemma-4-swift-mlx
Native Gemma 4 multimodal (text+vision+audio+video) inference for Apple Silicon via MLX Swift
- ltx-video-swift-mlx
LTX-2 text-to-video generation for Apple Silicon using MLX Swift
- yue2-mlx-swift
Swift/MLX port of YuE2 (m-a-p/YuE2-3B + YuE2-Vae) for Apple Silicon: lyrics + style -> song, on Mac and iPhone. Powers PocketAnthem on the App Store.
- swift-mlx-profiler
MLX performance profiling for Apple Silicon — GPU/CPU utilization, Chrome Trace (Perfetto), benchmarks
- mlx-voxtral-swift
Native Swift implementation of Voxtral speech-to-text for Apple Silicon with MLX
The full catalogue: 22 repositories, 27 models, datasets and demos
Support
The ports are free to use and three of the four apps are free. If one of them helped you, a coffee helps me keep going.



