# PocketPal AI

A free, open-source mobile app that runs language models entirely on-device, so chats, voice and tool use never leave the phone.

PocketPal AI is a private AI assistant that runs entirely on a phone, letting users chat with language models, give them a voice, and let them use tools, all on-device with no account or internet connection required after a model is downloaded. Every prompt, response and document stays on the device, and the only data that can leave it is what a user explicitly opts to share, such as leaderboard benchmark results.

It runs GGUF language models such as Gemma, Qwen, Phi and Llama fully offline, adds on-device neural text-to-speech, and supports "Pals" -- personalized assistants with their own model, system prompt and personality that can call built-in tools such as a calculator or date/time lookup inside a tool-use loop. Models, including gated ones, can be searched for and downloaded directly from the Hugging Face Hub.

It uses hardware acceleration paths for CPU, GPU (Metal on iOS, OpenCL/Adreno on Android) and NPU (Qualcomm Hexagon) with fallback, includes a benchmarking mode for tokens per second and memory use, and is MIT-licensed.

Website: https://github.com/a-ghorbani/pocketpal-ai