Favicon of Atomic Chat

Atomic Chat

A free local AI chat app with offline models and an OpenAI-compatible API for coding assistants, available on desktop and mobile.

Screenshot of Atomic Chat website

Atomic Chat is a local LLM app for people who want to chat with models on their own device or use them with coding assistants and AI agents. It began as a fork of Jan by Menlo Research.

  • Local model choice: Browse Hugging Face models such as Llama, Qwen, DeepSeek, Mistral and Gemma, with support for GGUF, MLX and ONNX formats.
  • Chat organization: Projects group conversations, persistent memory carries context across sessions, and custom assistants have their own system prompts.
  • Agents and tools: An OpenAI-compatible API lets compatible apps use local models, with integrations including Cline, Goose, OpenCode and OpenHands; MCP connections provide access to tools such as files and web search.
  • Code previews: A preview panel displays generated HTML, CSS and JavaScript.
  • Inference efficiency: TurboQuant reduces context-cache memory use, while supported models can use speculative decoding for faster generation.

The free app runs on Apple Silicon Macs, Windows x64 and Linux x86_64, with mobile apps for iOS and Android. Local chats work offline and stay on the device. Linux supports GGUF models only. Desktop inference uses llama.cpp with CPU execution or CUDA/Vulkan GPU acceleration, plus an MLX backend for Apple Silicon. RAM recommendations range from 8 GB for 3B models to 32 GB for 13B models.

The local API accepts connections from the same machine by default. Optional cloud-provider connections, including OpenAI and Anthropic, require your own API keys and send requests to the selected provider.

Similar to Atomic Chat