Documentation & Architecture.

Technical guides, runtime specifications, and architecture references for OfflineGPT on Android.

Features & On-Device Models

Explore Multimodal Vision, Visible Thinking, Snapdragon NPU Stable Diffusion, Whisper Voice, and the complete GGUF / LiteRT model catalog.

Learn how CodeLab's multi-file workspace, autonomous AI preview verification, virtual HTTPS WebView sandbox, and .ZIP exports work.

Review how On-Device mode keeps 100% of data on your phone, and how Hosted AI & BYOK operate without saving chats or training models.

Check recommended RAM tiers (4 GB, 6–8 GB, 8+ GB Snapdragon NPU) and follow the 3-step setup guide for Google Play.

ON-DEVICE INFERENCE RUNTIMES

GGUF Runtime (llama.cpp) 4-bit & 8-bit quantized LLMs (Qwen3, Llama 3.2, DeepSeek R1, Gemma 3n) with KV-cache prefix reuse

Google LiteRT (.litertlm) Hardware-accelerated multimodal vision & reasoning for Gemma 4 (E2B / E4B)

Diffusion Engine (QNN & CPU) Snapdragon Hexagon NPU (≤ 0.5s/step) & ARM CPU Stable Diffusion 1.5 / Anything V5

Speech Pipeline Offline whisper.cpp voice dictation + Supertonic 3 ONNX neural TTS (31 languages, 10 voices)

BYOK & CUSTOM ENDPOINTS

Supported Cloud APIs OpenAI, Anthropic (Claude), Google Gemini, DeepSeek, Groq, Mistral, xAI, Cohere, OpenRouter

Self-Hosted LAN / VPN Any OpenAI-compatible /v1/chat/completions endpoint (Ollama, LM Studio, vLLM, llama-server)

Key Storage Stored locally inside Android's encrypted application sandbox — never proxied through CodeAlp

Web Search Providers Zero-config DuckDuckGo Lite HTML parser + optional Brave Search & Tavily API keys

OfflineGPT on Google Play.

OfflineGPT · offlinegpt.tech · Developed by CodeAlp · com.offlinegpt.app