Local LLM Chat Support
Local LLM Chat downloads compatible MLX models from Hugging Face to your iPhone. Chat inference runs on the device.
Getting started
- Choose a public model on the Models screen and download it over a stable connection such as Wi-Fi.
- The initial download needs roughly 300–350 MB of storage per recommended model. Keep additional room for temporary processing beyond the displayed requirement.
- After saving finishes, select the model in Chat. A network connection is not required to run an installed model.
If a download does not finish
- Check the connection and available storage, then resume from the Models screen.
- For a model requiring authentication or license acceptance, review the distributor's terms. The recommended review model does not require a Hugging Face account.
- GGUF-only repositories are unsupported. Choose an MLX safetensors model.
- If the issue continues, delete the partial files and download the model again.
If generation stops or is slow
- Check memory, available storage, and thermal state in Device Monitor.
- Close other apps and wait for the device to cool before retrying if it is hot.
- Choose a smaller model and reduce maximum output or context length.
- Comparison runs two or three models sequentially to protect memory. If it stops, check the selected models and device state before retrying.
Delete history and data
Delete conversations from History, downloaded models from Models, and the Hugging Face access token from Settings. Uninstalling the app removes local conversation history and model files.
Contact
If the problem continues, email info@caen.co.jp with your device model, iOS version, model repository ID, and the displayed error. Do not send an access token or chat content.
FAQ
Is my chat sent outside the device?
No. Chat prompts and generated responses are processed on your iPhone. Network access to Hugging Face is limited to checking model information and downloading model files.
Does it work in Airplane Mode?
After a model finishes downloading, chat with that installed model and history viewing work offline. Checking or downloading a new model requires a connection.
Contact
For questions about Local LLM Chat, please contact CAEN Inc. by email.