PocketPal
PocketPal gives users the power of ChatGPT at their fingertips, allowing them to send a text and get a response in less than a minute.
Last verified:
What is PocketPal?
PocketPal is a mobile application that runs large language models (LLMs) directly on your phone without requiring internet connectivity, servers, or external data transmission. The app brings AI language models from Meta, Google, Alibaba, and other providers directly to Android and iOS devices, enabling users to chat privately with AI without any data leaving their device. It is built on llama.cpp and provides real AI inference on mobile hardware rather than acting as a cloud wrapper.
Key features include access to hundreds of open-source models from HuggingFace that can be downloaded directly to your device, complete offline operation in airplane mode or anywhere without signal, zero data transmission and zero telemetry, full control over model parameters like temperature and context length, built-in benchmarking to test model performance on your specific hardware, Metal acceleration on iOS and NNAPI on Android, and Pals – AI personas from Palshub.ai that give models personality and purpose. The app supports GGUF format models and includes Hugging Face integration for browsing and downloading models.
PocketPal is designed for people who stopped trusting the cloud with their conversations, those who want AI that works on planes or in tunnels off the grid, and anyone who cares about what runs on their own hardware. It requires modern phones with 6GB+ RAM for smaller models and 8GB+ RAM for better models, working best with flagship and recent mid-range phones from the last 2-3 years. The app is open source with every line of code public on GitHub for verification.
PocketPal pricing
Pricing model: Free
The app is completely free to use with no in-app purchases or advertisements. It is available for free download on Google Play Store for Android and App Store for iOS. There are no paid plans or subscription fees.
PocketPal pros
- Runs LLMs directly on phone without internet
- Zero data leaves your device ever
- No servers or cloud dependency
- Complete offline operation in airplane mode
- Hundreds of open-source models from HuggingFace
- Full control over temperature and context length
- Built-in benchmarking for your hardware
- Metal acceleration on iOS devices
- NNAPI acceleration on Android devices
- Support for GGUF format models
- Hugging Face integration built in
- Pals feature for AI personas with personality
- Open source with verifiable code
- Edit messages and retry AI generation
- Background downloads on iOS
- Screen stays awake during inference
- Multi-device support including iPad
- Localization support in Japanese and Chinese
- Access gated models with HF token authentication
PocketPal cons
- Requires 6GB+ RAM minimum for small models
- Needs 8GB+ RAM for good model performance
- Older or budget devices won't work well
- Smaller models than cloud ChatGPT
- Real trade-off in model size vs cloud AI
- Italic formatting with asterisks may not render
- No LaTeX formatting support yet
- AI response editing not implemented
Frequently asked questions about PocketPal
What is PocketPal AI?
PocketPal AI is a pocket-sized AI assistant powered by small language models that run directly on your phone. Designed for both iOS and Android, it lets you interact with various SLMs without internet connection. All processing happens entirely on-device, keeping your conversations, prompts, and data private without storage on external servers.
Does PocketPal require internet connection?
No, PocketPal provides complete offline operation. You can use it in airplane mode, without signal, underground, or anywhere else. The only internet needed is initially to download model weights from HuggingFace. After setup, you can chat offline anytime, anywhere.
What phone specifications does PocketPal need?
You need modern phones with 6GB+ RAM for smaller models and 8GB+ RAM for better models. Flagship and recent mid-range phones from the last 2-3 years run models well. Older or budget devices won't cut it because this is real inference, not a cloud wrapper.
Is PocketPal AI open source?
Yes, every line of code is public on GitHub. The privacy claims are verifiable through the source code. The project is licensed under MIT License and has 5k stars on GitHub with 463 forks.
What models can I use with PocketPal?
You can download hundreds of open-source models from HuggingFace including Danube 2 and 3, Phi, Gemma 2, Qwen, Llama models from Meta, and models from Google and Alibaba. The app supports GGUF format models optimized for ARM architecture used in mobile phones.
How do I download models in PocketPal?
Open the app, tap the Menu icon, navigate to Models page, choose a model and tap Download. Or tap the + button to add models from Hugging Face. You can search GGUF models directly on HF and select any quantization that fits your device's memory and storage.
What is the Pals feature?
Pals are AI personas from Palshub.ai that give models personality and purpose. You can create personalized AI assistants with different personalities and settings. There are two types: Assistant Pal with default model and system prompt, and Roleplay Pal with additional settings for location, AI role, and contextual parameters.
How does benchmarking work in PocketPal?
Navigate to the Benchmarking page to run performance tests comparing speed and efficiency. View detailed metrics like tokens per second and memory usage during AI response generation. You can share results and compare with other devices on the PocketPal AI Phone Leaderboard.
Can I access gated models on Hugging Face?
Yes, starting from v1.9.0 you can access gated models from Hugging Face with your authentication token. Navigate to Settings, tap Set Token, paste your personal access token from your HF account, and save. This enables access to both public and gated models.
Does PocketPal work on iPad?
Yes, starting from v1.8.12 in March 2025, PocketPal AI has full iPad support including landscape orientation. The app is optimized for both phones and tablets, providing multi-device support across the iOS and Android platforms.