Your private AI companion — fully offline.
Gemma 4 · Qwen 3.5 · on-device inference · no internet, no account, no cloud.
Download · Website · Report an issue · 中文文档
Natural disasters, remote expeditions, and everyday emergencies share one thing in common: connectivity fails exactly when you need it most.
Offline AI Assistant puts a capable large language model directly on your phone. Ask questions, get calm step-by-step guidance, analyze a photo, or run your own local API — with no Wi-Fi, no mobile data, and no cloud. Every conversation stays on your device.
An offline emergency & privacy AI companion supporting the latest local models — including Gemma 4 and Qwen 3.5. Reliable anywhere, truly private.
- Fully functional in airplane mode — no network required at any point.
- Private by design: all inference runs on-device. No account, no sign-up, no telemetry.
- We collect nothing and share nothing. Your chats never leave your phone.
- Privacy Mode can keep a conversation entirely in memory, leaving no history behind.
- Gemma 4 — Google's flagship open model, fast and capable, with image understanding.
- Qwen 3.5 — compact and quick for everyday questions.
- Import your own
.litertlmmodels at any time. - Use system AI where available (Gemini Nano on Android, Apple Foundation Models on iOS).
- Switch models anytime to balance speed, quality, and battery.
12 expert-crafted scenarios across urban disasters, extreme weather, and wilderness survival. Tap a scenario for calm, actionable, step-by-step guidance — even with zero signal. (See the full list.)
Just ask. "How do I treat a sprained ankle?" or "How can I filter water in the wild?" gets immediate, practical answers tuned for real situations — with streaming replies that appear as they are generated.
Attach a photo and ask about it: read a document, identify an object, or get advice from an image. On-device and offline.
Turn your phone into a private AI endpoint. The built-in OpenAI-compatible HTTP server lets tools like ChatBox or Open WebUI use your local model over your own Wi-Fi — no cloud, no API keys, no cost. (See the API examples.)
- 11 languages out of the box.
- Model downloads with resume and integrity verification.
- Fast streaming responses; dark and light themes.
- Works on a plane, off-grid, or whenever you simply don't want to be tracked.
| Model | Where it runs | Highlights |
|---|---|---|
| Gemma 4 (E2B) | On-device download | Flagship on-device model · multimodal image input |
| Qwen 3.5 | On-device download | Compact, fast everyday chat |
| Built-in model | Ships with the app | Instant start, zero-download setup |
| System AI | Android / iOS | Gemini Nano (Android), Apple Foundation Models (iOS) |
Custom .litertlm |
Import from file | Bring your own model |
Some distribution channels ship without a bundled model to keep the download small — the app then offers a one-tap download on first launch.
| Category | Scenarios |
|---|---|
| 🏙️ Urban Emergency Response | Residential Fire · Earthquake Safety · Citywide Blackout · Building Collapse |
| 🌪️ Extreme Weather | Flash Flood Response · Trapped in a Blizzard · Heatwave Readiness · Typhoon Preparedness |
| 🏕️ Wilderness Survival | Lost in the Desert · Stranded on an Island · Injured in the Mountains · Lost & Disoriented |
The Wilderness Survival pack unlocks with a one-time purchase — no subscription.
| Platform | Get it |
|---|---|
| 🤖 Android | Official download page · Google Play |
| 🍎 iOS | Official download page · App Store |
Additional mirrors and backup links are available on the official download page.
Requirements
- Android 11+ (API 30) on a 64-bit ARM device
- iOS 16.0 or later
The first launch may download a model depending on your build. After that, the app works entirely offline.
A built-in HTTP service lets any OpenAI-compatible client use your local model. The default endpoint is:
http://<device-ip>:8080/v1/chat/completions
Start the server from the in-app settings, then call it from another device on the same Wi-Fi.
curl http://<device-ip>:8080/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "gemma",
"messages": [
{"role": "system", "content": "You are an emergency rescue assistant."},
{"role": "user", "content": "How do I check if a building is safe after an earthquake?"}
]
}'curl http://<device-ip>:8080/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "gemma",
"stream": true,
"messages": [
{"role": "user", "content": "Hello"}
]
}'import OpenAI from "openai";
const client = new OpenAI({
apiKey: "local", // No validation for the local service — placeholder only.
baseURL: "http://<device-ip>:8080/v1",
});
const completion = await client.chat.completions.create({
model: "gemma",
messages: [
{ role: "user", content: "How do I purify drinking water during a flood?" },
],
});
console.log(completion.choices[0].message.content);- Replace
<device-ip>with your device's actual IP address. - Streaming vs. non-streaming is controlled by the
stream: trueparameter. - No real API key is needed — any placeholder works for the local service.
- Point clients such as ChatBox or Open WebUI at the same base URL to use your phone as their model backend.
The interface is available in 11 languages:
| English | 简体中文 | 日本語 | 한국어 |
| Deutsch | Français | Русский | Español |
| हिन्दी | العربية | Português |
Does it really work offline? Yes. Once a model is on your device, inference runs entirely on-device. No Wi-Fi or mobile data is required — airplane mode works.
Do I need an account? No. There is no sign-up, no login, and no tracking of any kind.
Which models can I use?
Gemma 4, Qwen 3.5, a built-in model, system AI where available, and any custom .litertlm model you import.
Where are my conversations stored? On your device only (local database). Use Privacy Mode to keep a conversation in memory with no history saved.
Can other apps use my phone's model?
Yes. Enable the OpenAI-compatible server and point any compatible client at http://<device-ip>:8080/v1 on the same network.
The model failed to load — what do I do? Open Settings, re-install the model, and export the logs for support. Switching the compute backend (CPU/GPU) can also help on some devices.
This repository is the public home for documentation and issue tracking.
- 🐛 Found a bug or have a question? Open an issue.
- ✨ Feature requests are welcome — please describe the use case.
- When reporting a problem, include your app version, device model, OS version, the model in use, and any exported logs from the app settings. This makes issues far quicker to resolve.
Offline AI Assistant provides informational guidance only and is not a substitute for professional emergency, medical, or rescue services. In a real emergency, always contact local authorities and trained professionals first.
The application itself is proprietary software. This repository hosts public documentation, API examples, and the issue tracker — the application source code is not published here.