Kancil is a private AI assistant that runs entirely on your Android device. No account. No cloud. No data ever leaves your phone.
Powered by Google's Gemma model running via llama.cpp, Kancil performs all AI inference locally — your conversations stay on your device, always.
FEATURES
• On-device AI — full LLM inference with no server required
• Vision — attach photos and ask questions about images
• Web search — optional DuckDuckGo search to ground answers in current information
• Streaming replies — see the response as it generates, token by token
• Background service — the model stays loaded and ready between chats
HOW IT WORKS
On first launch, Kancil downloads the Gemma model (~4.5 GB) from Hugging Face over Wi-Fi. After that, everything runs offline. The foreground service keeps the model in memory so responses start instantly.
PRIVACY
Kancil collects no personal data. It has no account system, no analytics, and no telemetry. The only network requests it makes are:
• Downloading the model on first launch (Hugging Face)
• DuckDuckGo search queries, only when you enable web search
All AI inference is local. Your chat history never leaves your device.
REQUIREMENTS
• Android 8.0 (Oreo) or higher
• ~5 GB free storage for the model
• 6 GB+ RAM recommended for best performance
• Internet connection for first-time model download