LlamaChat

LlamaChat sits at the intersection of two ideas: the appeal of large language models and the desire not to send your conversations to someone else's server. The app runs entirely on your Mac, loading model weights from local files and generating responses on your own hardware. Nothing about your prompts leaves the machine, which makes it a practical choice for people handling sensitive text, working offline, or simply preferring not to depend on a hosted service.

Practically, LlamaChat handles the tedious part of running local models for you. It can import raw PyTorch checkpoints in the format the community publishes them, or accept pre-converted .ggml files if you already have them, so you are not stuck doing conversion work in a terminal. Under the hood it leans on established open-source projects — llama.cpp and llama.swift — rather than reinventing inference. The result is a native Mac interface with model management, conversation history and a chat view, wrapped around a stack that the local-LLM community already trusts.

It is aimed at macOS users with an interest in self-hosted AI: developers experimenting with quantised weights, privacy-conscious writers, and hobbyists who want to try Alpaca, Vicuna or GPT4All models without touching a command line. The project is distributed as a free download and via Homebrew, requires macOS 13 or later, and runs on both Intel processors and Apple Silicon. It ships without any model files, so you source weights separately and are responsible for complying with their licences. The developer states the app is independent and not affiliated with Meta, Stanford or Nomic AI.