Atomic Chat
A free local AI chat app for macOS, Windows, Linux, iOS and Android, with offline models and an OpenAI-compatible API for coding assistants and agents.
Atomic Chat is a free local LLM app that also supplies models to coding assistants and AI agents. It builds on Jan by Menlo Research and runs on Apple Silicon Macs, Windows and Linux, with apps for iOS and Android. Local chats work offline without an account, and their data stays on your device.
You can browse Hugging Face models, including Llama, Qwen, DeepSeek, Mistral and Gemma. The app supports GGUF, MLX and ONNX formats; Linux runs GGUF models only. Chats and projects keep conversations organized, while persistent memory carries context across sessions. Custom assistants let you give different conversations their own system prompts, and a preview panel displays generated HTML, CSS and JavaScript.
Its OpenAI-compatible API lets other software use the models running on your computer. Integrations include Cline, Goose, OpenCode and OpenHands, and MCP connections give agents access to tools such as files and web search. The local server accepts connections only from your own machine by default.
Desktop inference uses llama.cpp and an Apple Silicon MLX backend. CPU execution and CUDA or Vulkan GPU acceleration are available through llama.cpp. TurboQuant reduces memory used by the model's context cache, and supported models can use speculative decoding to increase generation speed. RAM recommendations range from 8 GB for 3B models to 32 GB for 13B models.
A cloud option is also available. The app can connect to providers such as OpenAI and Anthropic with your own API keys; those requests go to the selected provider.
