pocketpal-ai: Run Language Models on Your Phone

Overview

PocketPal AI brings language-model chat to iOS and Android phones, running models locally rather than sending conversations to a cloud service. After downloading a model, users can chat offline, without an account.

It is aimed at people who value privacy and want to experiment with local AI on hardware they already own. The app also supports configurable assistants, voice output, and tools, though performance depends on the phone and chosen model.

Key Features

• Run GGUF language models locally through llama.cpp, with CPU, GPU, and supported Qualcomm NPU paths.
• Download and manage models, including searching Hugging Face for GGUF files.
• Create assistant and roleplay personas with their own model and system prompt.
• Use on-device text-to-speech engines for spoken responses.
• Let compatible assistants call built-in tools for calculations, date and time, and HTML rendering.
• Benchmark model speed and memory use, with an option to share results to a public leaderboard.
• Store chat history and settings on the device; leaderboard results and submitted feedback are the stated exceptions to its local-data approach.

Use Cases

• Privacy-conscious phone users who want to draft, brainstorm, or ask questions without sending prompts to an AI service.
• Travelers or people with unreliable connectivity who can download a model beforehand and chat offline.
• Mobile developers exploring local inference, React Native native modules, or tool-use flows on actual devices.
• Users who want a personalized assistant with voice output, without relying on cloud inference.

Getting Started

Install PocketPal from the App Store or Google Play, then download a model in the app. For development, see the README for setup details.

Alternatives

• cherry-studio: Cherry Studio is a desktop client for local and cloud models, while PocketPal AI is a mobile app focused on on-device inference.
• Open-LLM-VTuber: Open-LLM-VTuber centers on voice conversations with a Live2D avatar and supports hosted models, unlike PocketPal AI's mobile-first chat experience.

Considerations

Local inference avoids cloud dependence, but the available models and practical performance are constrained by a phone's memory, storage, and processing hardware. Model files must be downloaded first, and some hardware acceleration paths are device-specific with fallback behavior. Building from source also requires the relevant iOS or Android toolchain, in addition to Node.js and Ruby.

The repository is active, with regular commits and releases, and recent issue and pull-request handling. Its contributor base spans multiple people rather than depending primarily on one maintainer. That activity is encouraging, but the project still has a substantial open issue queue, so check current issues for platform-specific problems before relying on it. PalsHub and related account features require additional configuration, unlike core on-device chat.