Press kit

LlamaBox press kit.

Everything you need to write about, partner with, or evaluate LlamaBox — on one page.

A local AI layer connects several Android device and product forms.

Boilerplate

LlamaBox is a free Android app that runs large language models entirely on-device. It gives users private, offline AI chat with no cloud inference, no accounts, and no app analytics or conversation telemetry. Built with React Native and llama.cpp, LlamaBox is CPU by default; experimental acceleration may be available on supported devices so it works across the widest range of Android 7.0+ arm64 phones.

Key facts

  • Product: private offline AI chat for Android
  • Stack: React Native 0.81, llama.rn wrapping llama.cpp, GGUF models
  • Compute: CPU by default (ARM NEON, 4 threads); no accelerator required
  • Features: chat, vision, TTS readback, model hub, system monitor, offline history
  • Privacy: inference never leaves the device; no account required
  • License: AGPL-3.0 source release planned; commercial licensing available
  • Status: publicly available on Google Play; package com.llamabox
  • Creator: Aalhad (Mythos Labs)

What makes it different

Most "AI apps" send prompts to a vendor server. LlamaBox removes the server path entirely — the model runs on the phone. That makes it private by architecture, not by policy, and it works in airplane mode after the model file is downloaded.

Angles for coverage

  • Privacy / AI safety: an Android chat app that cannot leak prompts because it never sends them
  • Hardware access: CPU-default inference reaches mid-range and older phones, not just flagships
  • Offline / field use: journalists, travelers, students, healthcare workers, and defense edge cases
  • Open core: AGPL-3.0 source release planned plus commercial licensing for OEMs and enterprises
  • Comparison: how LlamaBox differs from PocketPal AI, MLC LLM, Ollama, and cloud assistants

Important caveats (please include)

  • LlamaBox is CPU by default; experimental acceleration may be available on supported devices; do not claim GPU acceleration.
  • The app source is not yet public; an AGPL-3.0 release is planned.
  • Phone CPUs are not datacenter GPUs — quality and speed are model-size dependent.

Links to cite

Media assets

Contacts

Available on Google Play

Try private offline AI on Android.

Install from Google Play, download a compatible model, then chat offline.

Get it on Google Play Release status GitHub