Prebuilt llama.cpp AAR for Android — on-device LLM inference (chat, completion, embeddings), no NDK, no cloud, no API key. GGUF models, arm64-v8a, API 24+.