GitHub Repository

M-Courtyard: Local AI Model Fine-tuning Assistant for Apple Silicon. Zero-code, zero-cloud, privacy-first desktop app powered by Tauri + React + mlx-lm.

137 starsTypeScript

Courtyard – Open-source macOS app for local MLX fine-tuning Text

Name: Courtyard – Open-source macOS app for local MLX fine-tuning Text
Availability: InStock
Author: tuwenbo0120

by tuwenbo0120·Feb 25, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●SolidSolve My ProblemShip It

Tauri GUI wrapper around mlx-lm—useful for Mac users, but local fine-tuning UIs already exist.

Strengths

•Unified pipeline eliminates Python CLI juggling, environment setup, and JSONL formatting headaches.
•Integrated A/B chat testing against base vs. fine-tuned adapter reduces manual validation work.
•Direct Ollama export with quantization removes separate conversion step.

Weaknesses

•macOS-only limits audience to Apple Silicon users; no Linux or Windows parity.
•Competing with Ollama's UI, LM Studio, and other local LLM GUIs in the same space.

Post Description

I've been building Courtyard, a macOS desktop app designed to make local LLM workflows on Apple Silicon less tedious.

The motivation: I was tired of juggling multiple Python CLI scripts, JSONL formatting, and environment issues just to run a simple LoRA fine-tune on my Mac.

Courtyard is essentially a UI wrapper around mlx-lm combined with data preparation tools. It handles:

Dataset formatting and cleaning (privacy filtering, deduplication). Local LoRA fine-tuning via MLX on Apple Silicon. An integrated chat UI for A/B testing the base model vs. the fine-tuned adapter. Exporting to GGUF or directly to an Ollama runtime. The stack is Tauri 2.x + React + Rust + Python (mlx-lm). It's fully open-source (AGPL).

Repo: https://github.com/Mcourtyard/m-courtyard

I'd love to hear your thoughts on the architecture, MLX implementation, or any edge cases you run into. Happy to answer technical questions.