Atomic Chat

A chat app that runs open AI models directly on your computer or phone. It's built for people who want the convenience of a hosted assistant without sending prompts to someone else's servers. It works as an open source alternative to LM Studio and similar local model apps.

Install it like any normal app, pick a model, and start chatting. No account is required, and it works fully offline.

• 1,000+ models: Llama, Qwen, DeepSeek, Mistral, Gemma and more, browsed from Hugging Face and downloaded in one click. GGUF, MLX and ONNX formats are supported.
• Private by design: your data stays on the device, with no message caps, subscription or rate limits.
• Local agents: create and run autonomous workflows on your machine, and get tools like Hermes, OpenClaw and Cline running with local models.
• TurboQuant built in: compresses the KV cache for faster inference and longer context windows, so larger models run more smoothly on modest hardware.
• Organized workspace: Chats and Projects keep contexts separate, with persistent memory across sessions.
• Every platform: macOS (Intel and Apple Silicon), Windows, Linux, iOS and Android.

The code is open, so you can inspect exactly what the app does. If you're weighing options, Jan and GPT4All cover similar ground. This one stands out for its built-in quantization work and its one-click agent support.