PocketPal setup: offline AI chat on Android and iPhone

This tutorial explains how to run models locally with PocketPal on Android and iPhone. The presenter demonstrates chat with internet access disabled and describes the app as open source. The walkthrough uses LD Player, an Android emulator, and covers installation through Google Play or the App Store.

Model choice depends on the phone's CPU, RAM and available storage. The presenter explains that stronger hardware can handle larger models and respond faster, while download sizes appear beside each model. Smart Chat suggests models for conversation; another section suggests models for coding. An initial model reportedly struggles with Persian, while the later Gemma chat demonstration produces Persian replies. The app interface also has a Persian language option.

The walkthrough covers downloading models, importing a file from phone storage, and adding models through Hugging Face. For the Hugging Face connection, the presenter instructs viewers to create an account and enter an access token. A benchmark section tests a selected model on the device, and advanced settings include CPU core allocation.

PocketPal also has a remote model option that connects to a server address. The presenter's privacy claims concern on-device execution. The final demonstration covers switching models, managing chat history and adjusting generation settings, including the maximum token count.