PocketPal AI Android

Your AI. No one else's.

PocketPal runs large language models directly on your phone. No servers. No internet. No data leaves your device. Ever.

This is AI you own — not AI you rent.

WHAT YOUR PHONE NEEDS

Modern phones are powerful enough to run real AI. PocketPal uses that power.

You'll need 6GB+ RAM for smaller models, 8GB+ for the good stuff. Flagship and recent mid-range phones from the last 2-3 years run models well. Older or budget devices won't cut it — this is real inference, not a cloud wrapper.

Not sure? Download a small model and benchmark it. The app will show you exactly what your device can handle.

WHAT YOU GET

- Hundreds of open-source models from HuggingFace, downloaded directly to your device
- Complete offline operation — airplane mode, no signal, underground, anywhere
- Zero data transmission. Zero telemetry. Verify it yourself: we're open source
- Full control over model parameters — temperature, context length, system prompts
- Built-in benchmarking to test model performance on your hardware
- Metal acceleration on iOS, NNAPI on Android
- Pals — AI personas from Palshub.ai that give models personality and purpose

WHAT THIS IS NOT

This is not ChatGPT. Cloud AI has trillion-parameter models running on server farms. Your phone runs smaller models locally. The trade-off is real — and it's one we're proud of.

You get something no cloud service can offer: conversations that exist only on your device, AI that works without internet, and complete freedom to run any model you choose. No content policies imposed by us. No behavioral guardrails we decided for you. Your rules.

WHO THIS IS FOR

People who stopped trusting the cloud with their conversations. People who want AI that works on a plane, in a tunnel, off the grid. People who care about what runs on their own hardware.

If you're looking for the "best" AI at any cost and don't care about privacy — use ChatGPT. It's great at what it does.

If you want AI that's yours, PocketPal is how you get it.

OPEN SOURCE

Every line of code is public on GitHub. Our privacy claims aren't marketing — they're verifiable. If you don't trust us, read the source. That's the point.

• llama.cpp capability discovery: PocketPal now detects a connected server's vision support and context size, and warns when a chat runs out of room.
• Internet search in chat: Pals can search the web and read pages. Bring your own key (Brave, Tavily, Exa).
• Speculative decoding (experimental): pair a small draft model for faster replies.
• Pin chats, export chats as Markdown.
• Polish and Portuguese (PT), plus a searchable language picker.
• Added Spark2.5 model support