Local LLM Scripts

A collection of local LLM-powered productivity tools that run as Raycast commands. All inference runs on-device via MLX — no API keys, no data leaving your machine.

Requires Apple Silicon (M1 or later).

---

Scripts

Script	Raycast Command	What it does
ask.py	ask	General-purpose Q&A — plain text answers, no Markdown
fixtext.py	fixtext	Fixes grammar, spelling, and clarity while preserving your voice
fixmessage.py	fixmessage	Rewrites a work message to be clearer and more professional
fixprompt.py	fixprompt	Optimizes a rough prompt into a structured LLM prompt

fixtext, fixmessage, and fixprompt automatically copy the result to your clipboard.

---

Setup

1. Clone and install dependencies

git clone https://github.com/alexspecter/Local-LLM-Scripts
cd Local-LLM-Scripts
uv sync

Requires uv. Install with brew install uv or pip install uv.

2. Configure your model

cp .env.example .env

Edit .env and set MODEL_PATH to any MLX-compatible model from Hugging Face:

MODEL_PATH=mlx-community/Qwen2.5-32B-Instruct-4bit

The model will be downloaded automatically on first run. A 4-bit quantized 32B model requires ~20GB of RAM.

3. Add scripts to Raycast

1. Open Raycast → Settings → Extensions → Script Commands
2. Add a new script directory pointing to the scripts/ folder in this repo
3. The scripts will appear as commands in Raycast

---

Dependencies

• mlx-lm — local LLM inference on Apple Silicon
• promptloop — prompt utilities
• python-dotenv — .env config loading