Atomic Chat

Atomic Chat is a free, open source app for chatting with AI models that run entirely on the user's own device. It downloads and runs open models locally, so conversations never leave the machine, work fully offline and carry no rate limits or subscription. Desktop builds are available for macOS 13 and later on Apple Silicon, Windows x64 and Linux x86_64, with mobile apps on iOS and Android and an option to run in the cloud. It is for people who want private AI without a cloud account, from individuals to developers experimenting with open models. A built in browser lists more than 1,000 models from Hugging Face, including Llama, Qwen, DeepSeek, Mistral, Gemma, Kimi and MiniMax, installed with one click in GGUF, MLX or ONNX format. It includes TurboQuant, a KV cache compression technique the site says cuts memory use by at least six times without quality loss, allowing longer context windows and larger models on the same hardware. Chats and Projects keep work organised, memory persists across sessions, and one click launchers start agent tools such as Hermes, OpenClaw and Cline on top of Atomic Chat models. The app is free, with no paid tier listed on the site, and the source code is open for inspection. Atomic Chat shares its terminal installer with Atomic Agent, a local operator agent from the same team, so users who want autonomous task execution rather than chat can move between the two.