Repolode
Submit
a-ghorbani

a-ghorbani/pocketpal-ai

A React Native mobile app that runs language models, text-to-speech and tool use entirely on your phone, offline and without an account.

Open-source alternative to:ChatGPTGoogle AI Edge Gallery

View on GitHub
Stars
8,732
Forks
967
Open issues
132
Last commit
13 hours ago

What is it?

PocketPal AI is a mobile app for iOS, iPadOS and Android that runs GGUF language models directly on the device. Chats, prompts and documents stay on the phone, and the app works without an internet connection or account once a model is downloaded.

Inside the app, users can create customized assistants called Pals with their own model, system prompt and personality, and browse community Pals on PalsHub. Pals can call built-in tools such as a calculator, a date/time lookup and an HTML renderer. On-device text-to-speech with engines like Kokoro gives assistants a voice, and models can be searched and downloaded from the Hugging Face Hub, including gated ones with an access token.

The stack is built on React Native with llama.rn bridging llama.cpp for LLM inference and ONNX Runtime for TTS. Inference can use CPU, Metal on iOS, OpenCL on Qualcomm Adreno GPUs, or Qualcomm Hexagon NPUs, with fallback when a backend is unavailable. The project is MIT-licensed and includes a benchmarking page for measuring tokens per second and memory use.

Key features

  • 01Runs GGUF language models fully offline on iOS and Android
  • 02On-device text-to-speech with engines such as Kokoro
  • 03Custom Pals with their own model, system prompt and persona
  • 04Built-in tools for calculation, date/time and HTML rendering
  • 05Downloads models from the Hugging Face Hub, including gated ones
  • 06Hardware acceleration via CPU, Metal, OpenCL and Qualcomm NPU

Use cases

  • Chatting with an AI assistant on a plane or somewhere without connectivity
  • Keeping sensitive notes or documents private by avoiding cloud processing
  • Testing and comparing how different GGUF models perform on a specific phone
  • Building and sharing custom roleplay or task-specific assistants

Quick start

  1. 1

    Install the app from the App Store or Google Play

  2. 2

    Clone the repository

    git clone https://github.com/a-ghorbani/pocketpal-ai
  3. 3

    Install JS dependencies

    yarn install
  4. 4

    Run on Android emulator

    yarn android

Things to know

  • Developer setup needs Xcode, CocoaPods, Android Studio and a pinned Node version.
  • Models must be downloaded to the phone before chatting, and larger models need more memory.

Topics

#on-device-ai#llm#react-native#llama-cpp#offline#text-to-speech#mobile#gguf

FAQ

What is pocketpal-ai?
A React Native mobile app that runs language models, text-to-speech and tool use entirely on your phone, offline and without an account. PocketPal AI is a mobile app for iOS, iPadOS and Android that runs GGUF language models directly on the device. Chats, prompts and documents stay on the phone, and the app works without an internet connection or account once a model is downloaded.
Is pocketpal-ai free to use? What license does it have?
pocketpal-ai is open source under the MIT license. Check the license terms before using it in commercial products.
Is pocketpal-ai still maintained?
Actively maintained. The last commit to pocketpal-ai was on Oct 20, 2026.
What can I use pocketpal-ai for?
Chatting with an AI assistant on a plane or somewhere without connectivity Keeping sensitive notes or documents private by avoiding cloud processing Testing and comparing how different GGUF models perform on a specific phone Building and sharing custom roleplay or task-specific assistants
How do I install pocketpal-ai?
Install the app from the App Store or Google Play Clone the repository: git clone https://github.com/a-ghorbani/pocketpal-ai Install JS dependencies: yarn install Run on Android emulator: yarn android
Where can I get pocketpal-ai?
The source code and installation instructions are on GitHub: https://github.com/a-ghorbani/pocketpal-ai

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects