1chat is an iOS app that lets you chat with the DeepSeek-R1 model locally.
Designed with privacy and security in mind, 1chat requires no internet connection or special permissions. It runs entirely on-device with an integrated model, ready to use right out of the box.
1chat is based on the open-source project fullmoon-ios, with custom development and integration of the DeepSeek-R1-Distill-Qwen-1.5B model.
It is fully compatible with DeepSeek R1 and supports deep thinking (DeepThink).
You can download 1chat from App Store
1chat can run on iPhone, iPad, Mac, and Vision Pro.
- Built-in DeepSeek-R1 model(qwen-q4-1.5b), ready to use.
- Runs entirely offline—no internet required.
- Fully adapted for the R1 model.
- Supports multiple conversation management.
- Optimized for Apple chips.
- Removed the system prompt(temporarily).
- Localize for Chinese.
- Removed the model downloader.
- Currently supports only DeepSeek-R1 (temporarily).
git clone https://github.com/OJZen/1chat.git
Download DeepSeek-R1 model files from huggingface
Alternatively, you can use another model, but it must meet these requirements:
- It must be a DeepSeek-R1 distill model.
- It must be in MLX format.
The 1.5B model work best on my iPhone 15 pro -- it's fast enough for smooth performance.
Place all model files into the model folder.
Open the 1chat project in Xcode, then build and run.
DeepSeek-R1 currently has an annoying issue where the output sometimes repeats endlessly.
I don’t know how to fix it, so for now, I’ve added a method that automatically truncates the output when repetition is detected.