br-AI-n: Offline Private AI

br-AI-n is a private, offline-first AI assistant that runs entirely on your Android phone. Download leading open-weight language models and chat with them directly on your device — no cloud AI, no remote servers, no account.

Because everything runs locally, your chats, voice notes, and images never leave your phone. When you need current information, optionally switch on Web Search to get live links and citations — still without sending your conversations to an AI service.

🔒 Private by default
• 100% on-device inference: no cloud AI, no tracking, no analytics.
• Works in airplane mode once a model is downloaded.
• Chat history is stored in a local database on your phone.

🌐 Optional Web Search with citations
• Off by default. Enable it to fetch live links shown as numbered citations with the source name.
• Returns links only — it never opens, scrapes, or feeds web page content into the model.
• Sources include Wikipedia, DuckDuckGo, GDELT News, arXiv, Crossref, PubMed, Stack Overflow, Hacker News, Open Library and more.
• While Web Search is off, no web feature makes a network request.

🎭 Personas
• Built-in personas (Assistant, Researcher, Creative Writer, Code Expert, Joker, Influencer) plus custom personas with your own name, emoji and instructions.

🧠 Transparent reasoning
• See the model think in a collapsible reasoning panel, including multi-step web searches.

🎙️ Voice
• Dictate with your device's speech input and hear replies read aloud with low-latency streaming text-to-speech.

👁️ Vision & OCR
• Attach photos, receipts, diagrams or documents to transcribe and discuss with multimodal models.

📊 Live performance stats
• Tokens per second, context usage, free RAM and storage — so you can choose the right model.

📂 Local chat history
• Sessions saved on-device. Resume or delete anytime.

🤖 Model library
• One-tap downloads of optimized GGUF models: Llama 3.2, Qwen 2.5, Qwen3, Gemma 4, Phi-4-mini, DeepSeek-R1 Distill, LFM 2.5 VL and more, with a faster Q4_0 option where available.

🛡️ No accounts, no subscriptions
• No sign-in, no ads, no recurring fees.

⚙️ Powered by llama.cpp
• A mobile-optimized C++ engine runs quantized models efficiently on your phone's CPU.

Note: generation speed, loading time and memory use depend on your device's processor and RAM. The built-in Hardware Profile in the Model Manager recommends the best model size for your phone.