Local LLM

Offline on-device LLM chat for Android (llama.cpp GGUF + LiteRT-LM)

Local LLM Android is a privacy-first, fully offline on-device Large Language Model (LLM) application and server for Android. Powered by a hybrid dual-runtime architecture combining llama.cpp and Google LiteRT-LM, it executes state-of-the-art AI models directly on your device's CPU, GPU, and NPU—without requiring internet connectivity or cloud servers.

⚡ Key Highlights Real-Time Hardware Benchmarking: Displays prompt prefill speed (promptSpeed tokens/sec), generation decode speed (tps), and active context token counts on every message. Interactive Voice Mode: Hands-free conversational loop: On-device Speech-to-Text (STT) → LLM generation → Text-to-Speech (TTS) response. Visualized with an animated reactive Voice Orb. Streaming Reasoning State Machine (ReasoningStreamParser): Elegantly parses <think> ... </think> tags across token streaming chunks for reasoning models like DeepSeek-R1. Smart Out-of-Memory (OOM) Protection: Inspects ActivityManager.MemoryInfo before model allocation to prevent native memory exhaustion crashes (SIGSEGV). Resilient Mobile Downloader: Background foreground-service downloads with pause/resume support and integrity verification.

🔒 Enterprise-Grade Security & Privacy Hardware-Backed AES-256-GCM Encryption (ChatCrypto)

Cryptographic keys are generated and stored in the hardware Android KeyStore (Secure Element / TEE). Authenticated Encryption (AEAD) ensures conversational data cannot be intercepted or tampered with. Strict prevention of silent downgrades to insecure hardcoded keys. 100% Offline & Private

Zero telemetry, zero analytics, zero external API dependencies. Conversations and prompts remain strictly on your physical device in an encrypted Room SQLite database.