PocketPal AI

PocketPal AI runs language models on phones and tablets for people who want an assistant they can use without sending conversations to a server. Core chat works offline after a model download and doesn't require an account.

• Model choice: Run GGUF models such as Gemma, Qwen, Phi and Llama through llama.cpp, with models from local storage or Hugging Face, including gated models when you supply an access token.
• Spoken replies: Hear responses through on-device text-to-speech, including Kokoro voices, without cloud calls.
• Personal assistants: Create Pals with their own model, instructions and personality for assistant or roleplay conversations.
• Built-in tools: Compatible Pals can perform calculations, check dates and times, and render HTML within a conversation.
• Device benchmarks: Measure model speed and memory use on your hardware, with optional result sharing to the AI Phone Leaderboard.

The app is free and open source under the MIT license. It runs on iOS, iPadOS and Android, with CPU inference and acceleration through supported GPUs or Qualcomm Hexagon NPUs. It falls back to CPU when acceleration isn't available; model choice depends on your device's memory and storage.

Prompts, replies and documents stay on the device during local use. Model downloads need an internet connection. PalsHub is a separate online service for discovering community Pals, including paid ones, and isn't required for offline chat. Benchmark results leave the device only if you choose to share them, and submitted feedback also leaves the device.