vovw / kunzite

local, private, gui for running your inference engine of choice

Kunzite

Kunzite is a lightweight, bloat-free GUI application for local AI chat using the lugia inference engine (based on mlx).

Features

• Local AI: Run AI chat completions entirely on your device
• MLX Inference Engine: Utilize Apple's MLX framework for fast, efficient inference
• On-Device Inference: Ensure privacy and offline functionality
• Bloat-Free GUI: Minimalist interface for distraction-free work

Quick Start

1. Clone the repository
2. Install dependencies - imgui, lugia
3. Run make to build the project
4. Execute ./kunzite to start the application

• Kunzite uses a simple client-server architecture
• GUI: Implemented with Dear ImGui for a lightweight, cross-platform interface
• LLM Client: Handles communication with the local inference server
• MLX Server: Runs the language model using the MLX framework using lugia inference engine

License

This project is licensed under the MIT License