OfflineGPT

OfflineGPT

On-device AI for Android — chat, images, voice, TTS, and a Python sandbox for API models. OfflineGPT Core is free. Hosted models are optional.

Get it on Google PlayDownload on the App Store

iOS version: Offline AI Studio on the App Store

What OfflineGPT can do

Android 8.0+, 64-bit ARM. Models run on-device — CPU, GPU (Vulkan/OpenCL), or Snapdragon NPU.

Offline by default

Chats, generations, and voice input stay on your device by default. No ads, no tracking. Cloud processing happens only if you choose hosted models or your own API keys.

Python sandbox + network

For API models only: switch on Power Mode in the plus menu and the model can run Python in a sandboxed runtime with optional network access. Hit APIs, fetch data, automate workflows. Not available for downloaded on-device models.

Gemma 4 E4B — multimodal

Google's on-device model understands text, images, and audio in one conversation. Optional web search when you want it.

Thinking models

Qwen3 (0.6B – 4B) and SmolLM3-3B show step-by-step reasoning before answering. Good for problems that need a second look.

Web search, three engines

DuckDuckGo Lite needs no key and is the default. Brave and Tavily work with a key of your own. Available to API models and to the LiteRT models on your device.

Vision & image generation

Describe images with InternVL or SmolVLM2. Generate art with Stable Diffusion (Anything V5, AbsoluteReality, ChilloutMix, CuteYukiMix).

Snapdragon NPU acceleration

Image generation runs on the Hexagon NPU on supported Snapdragon devices — faster, cooler, less battery drain.

Whisper voice input

Dictate prompts in any language with on-device speech recognition. Compact and enhanced Whisper models are optional downloads — pick what you need.

Read aloud / TTS

Answers are read back via system voice or the optional Supertonic 3 neural engine — 31 languages, 10 voices, streaming during generation.

AI writing tools

Summarize, grammar, tone, simplify, social posts, document processing, email drafting, and Idea Blast brainstorming.

Code Lab

Prototype HTML, CSS, and JavaScript with a built-in live preview and an AI coding assistant alongside.

Offline OCR

Extract text from photos and PDFs on-device using ML Kit — no cloud OCR.

Custom AI tools

Build your own tools with a custom name, icon, system prompt, and optional file knowledge base. Pin them to the home screen.

Projects for your chats

Group conversations into project folders in the sidebar instead of scrolling one long list.

22 themes

Light and Dark are the default themes. Beyond them sit twenty atmospheric themes, all of them free.

On-device model catalog

Download sizes are approximate. Custom GGUF imports and the in-app Hugging Face browser are supported too.

Text models (GGUF · llama.cpp)

LFM2.5 1.2B

LiquidAI

~731 MB

Llama-3.2-1B

Meta

~1.4 GB

Gemma-2B

Google

~1.85 GB

Qwen2.5-3B

Alibaba

~2.1 GB

Rocket-3B

Community

~3.1 GB

Thinking models (GGUF · llama.cpp)

Qwen3 0.6B Thinking

Alibaba

~640 MB

Qwen3 1.7B Thinking

Alibaba

~1.8 GB

Qwen3 4B Thinking

Alibaba

~2.6 GB

SmolLM3-3B

Hugging Face

~1.9 GB

Multimodal — text + image + audio (LiteRT)

Gemma-4-E2B-it

Google (LiteRT)

~2.58 GB

Gemma-4-E4B-it

Google (LiteRT)

~3.66 GB

Text models (LiteRT)

Qwen2.5-1.5B-it

Alibaba (LiteRT)

~1.60 GB

Qwen3-4B-it

Alibaba (LiteRT)

~2.66 GB

Vision models — image understanding (MNN)

InternVL2.5 1B

OpenGVLab (MNN) · 4 GB RAM

~888 MB

SmolVLM2 2.2B

Hugging Face (MNN) · 6 GB RAM

~2.48 GB

Image generation (Stable Diffusion · MNN + NPU)

Anything V5

Anime — CPU or NPU

~1.25 / 1.0 GB

AbsoluteReality

Photorealistic — CPU or NPU

~1.25 / 1.0 GB

ChilloutMix

Photorealistic — CPU or NPU

~1.25 / 1.0 GB

CuteYukiMix

Anime — CPU or NPU

~1.25 / 1.0 GB

Voice input & text-to-speech

Whisper base (optional)

whisper.cpp

~78 MB

Whisper enhanced (optional)

whisper.cpp

~142 MB

Supertonic 3 (optional)

ONNX · 31 languages · 10 voices

~380 MB

OfflineGPT Core is free

Hosted models are an optional Google Play subscription.

Core

Local chat, image generation, voice, writing tools, OCR, and custom tools — included, without a feature paywall.

Hosted models

Four hosted models, including new GPT-6 Luna. Gemma 4 runs in Switzerland. CHF 14.90 per month in Switzerland, unlimited personal use. Coming soon on iPhone. No OfflineGPT account and no cloud chat archive.

Learn more about hosted models

Frequently asked questions

Do I need internet?

To download models and app updates, yes. Once downloaded, local chat, image generation, and voice input work fully offline. Optional: web search, API mode, and the Python sandbox (API models only) need a connection when used.

Is OfflineGPT free?

Yes. OfflineGPT Core is free: local chat, image generation, voice input, and writing tools. The optional hosted-models subscription adds four hosted models, including new GPT-6 Luna (CHF 14.90 per month in Switzerland; the Google Play price for your region applies). Gemma 4 runs in Switzerland.

What is the Python sandbox?

A sandboxed Python runtime for API-model chats only — when you use cloud providers with your own keys. Enable network access so scripts can call APIs or fetch remote data. It is not available for local on-device models (Gemma 4, GGUF downloads, etc.).

What is Gemma 4 E4B?

Google's multimodal on-device model — it understands text, images, and audio attachments in a single conversation. It runs via LiteRT-LM, not llama.cpp, and supports an optional web search tool you can toggle on or off.

Is there an iPhone or iPad version?

Yes. On the App Store the app is listed as Offline AI Studio — same offline-first philosophy, built for iOS with Core ML.

How does image generation work?

Stable Diffusion models (Anything V5, AbsoluteReality, ChilloutMix, CuteYukiMix) run locally via MNN. On supported Snapdragon devices, NPU variants run on the Hexagon DSP for faster generation.

How does voice input work?

Download a Whisper model in the app (base or enhanced), grant microphone access, tap the mic, and dictate. Audio is transcribed on-device — nothing leaves your phone.

What is text-to-speech?

Answers can be read aloud automatically or on tap. The system voice works with zero setup. Supertonic 3 is an optional ~380 MB download with 31 languages and 10 neural voices, and supports streaming TTS during generation.

How does privacy work?

By default, prompts and chat history stay on your device. Data only leaves your phone when you explicitly use optional features: model downloads, Gemma 4 web search, your own API keys, the Python sandbox with network, or OfflineGPT hosted models. Full privacy policy.

Get OfflineGPT

Free download on Google Play · Android 8.0+

Get it on Google PlayDownload on the App Store