Airux Pocket AI
A local LLM chat app for Android runs GGUF models offline, analyzes photos on-device and supports optional DuckDuckGo search.
Airux Pocket AI runs AI conversations on Android phones for people who want offline chat and photo analysis without cloud inference. You don't need an account, and downloaded models run directly on your phone.
- Model choice: Run GGUF models through llama.cpp, with a catalog that includes Qwen3, Llama 3.2, Gemma 3 and SmolVLM2, plus recommendations based on your phone's RAM.
- Photo analysis: Attach photos to a chat with a supported vision model and process images locally in the same conversation as text.
- Chat context and coding: Keep context within each conversation and use a coding mode to continue long code responses, with prompts tailored to the selected model.
- Optional web search: Qwen3 1.7B and Llama 3.2 3B can request DuckDuckGo searches without API keys or a cloud service managing the conversation; searches require internet access and send queries to DuckDuckGo.
The app requires an arm64 phone running Android 8 or later. Llama 3.2 1B and the smaller SmolVLM2 model require at least 4 GB of RAM; larger catalog choices require 6 or 8 GB. Photo analysis and search availability depend on the selected model.
Models download from Hugging Face over the internet, then support offline chat. Chat history and model files stay in the app's private storage. The app is open source under the MIT license.