PocketPal AI
A mobile AI assistant that runs GGUF models on iOS and Android. Core chat works offline after a model download and needs no account.
PocketPal AI is an open source assistant for people who want to run language models on a phone or tablet. It works on iOS, iPadOS and Android. Once you've downloaded a model, you can chat offline without an account, and your prompts, replies and documents stay on your device. The app is licensed under MIT.
It runs GGUF models such as Gemma, Qwen, Phi and Llama using llama.cpp. You can add a model from local storage or find one through Hugging Face, including gated models if you have an access token. Downloading models requires a connection; using a downloaded model does not. Inference can use the phone's CPU, GPU or a supported Qualcomm NPU, with CPU fallback. An in-app benchmark shows model speed and memory use on your device.
PocketPal also has on-device text-to-speech, including Kokoro voices. You can create Pals with their own model and instructions, including roleplay settings. Capable Pals can use built-in tools for calculations, date and time, and HTML rendering during a conversation. You can edit a message to generate a new response or retry with a different model.
PalsHub lets you find community-made Pals. It's separate from core offline chat. Benchmark results reach the AI Phone Leaderboard only if you choose to share them, and feedback leaves the device only when you submit it.
