OfflineLLM
A fully offline private AI chat app for Android that runs any GGUF model on-device through llama.cpp with no network permission at all, plus optional Vulkan GPU acceleration and per-device optimized CPU kernels.
OfflineLLM has no INTERNET permission declared in its manifest, so it cannot phone home, and all inference runs on-device through llama.cpp with runtime CPU dispatch that picks the best available kernel set, such as dotprod, fp16, i8mm or SVE, for the phone's specific chip. An opt-in Vulkan acceleration toggle can offload the entire model to the phone's GPU per layer, with automatic fallback to CPU, and multi-turn chat only feeds each new message into the KV cache instead of reprocessing the whole conversation each turn.
Users import any GGUF model file through the system file picker, and the app offers streaming token-by-token output, multiple renameable and searchable conversations, translation across 75-plus languages, adjustable sampling parameters, and several built-in system prompts such as coder, creative writer and tutor. Security features include encrypted settings, an optional biometric lock, secure file deletion, and a release build that verifies its own APK signing certificate at startup and refuses to run if the package has been repackaged.