Skip to content

Latest commit

 

History

History
68 lines (41 loc) · 2.62 KB

File metadata and controls

68 lines (41 loc) · 2.62 KB
layout default
title Quick Start
nav_order 2
description Download Off Grid and run your first local AI model in under 5 minutes - no account, no API key, no cloud.

Quick Start

Run your first local AI model in under 5 minutes. No account. No API key. No internet after setup.


Step 1 - Download Off Grid

iOS: Download on the App Store - requires iPhone 12 or newer (4GB RAM+)

Android: Get it on Google Play - requires Android 10+, 4GB RAM+

Or grab the latest APK directly from GitHub Releases.


Step 2 - Pick a model

When you open the app, you'll see the model picker. If you're unsure, start here:

You want Start with Size
Fast chat, 3–4GB RAM Qwen 3.5 0.8B ~0.8GB
Best for most phones Qwen 3.5 2B ~1.7GB
Best quality (8GB RAM) Qwen 3.5 9B ~5.5GB
Vision + reasoning Gemma 4 E2B ~1.5GB
Image generation SD 1.5 Palettized (iOS) / Absolute Reality (Android) ~1GB

Not sure? Pick Qwen 3.5 2B. It fits comfortably in 4GB RAM, supports 262K context, and is the best starting point for most phones.


Step 3 - Download and run

Tap a model → Download. This is the only time you need internet. The download goes to your device storage.

Once downloaded, tap Load - the model loads into RAM. On first load this takes 5–15 seconds depending on model size.

Type your first message. You're now running AI locally.


Step 4 - Go offline (optional)

Put your phone in airplane mode. Everything still works.


What's next

  • [Which model should I use?]({{ '/guides/which-model' | relative_url }}) - full comparison table by device and use case
  • [Connect your home Ollama server]({{ '/guides/ollama-android' | relative_url }}) - use bigger models from your desktop via LAN
  • [Run Stable Diffusion on Android]({{ '/guides/stable-diffusion-android' | relative_url }}) - generate images completely on-device

Community

Stuck, or want to share what you're building? Join the Slack community.

The app is open source - view it on GitHub.