You probably assume your iPhone keeps things private. Apple markets privacy as a core feature. But when you use ChatGPT, Gemini, or any cloud AI app on your iPhone, none of Apple's privacy protections matter. Your conversations leave your phone, land on someone else's server, and stay there.
Every prompt you type. Every personal question you ask. Every document you upload. It all gets logged, stored, and in many cases used to train the next version of the model. Apple's privacy architecture can't protect you from data you voluntarily hand to a third party.
There's a way to use AI on your iPhone where nothing ever leaves. No servers, no accounts, no tracking. Everything runs on your phone's own processor.
What Cloud AI Services Actually Do With Your iPhone Data
Even on an iPhone with all privacy settings enabled, here's what happens when you use cloud AI:
ChatGPT on iPhone: Your conversations go to OpenAI's servers. By default, your chats can be used to train future models. You can opt out, but it's buried in settings. Even opted out, OpenAI retains conversations for 30 days. Human reviewers may read your chats for safety monitoring.
Gemini: Google saves your conversation history for 18 months by default. Conversations flagged for human review are stored separately for up to 3 years. Your chats are used to "improve the performance" of Google's services.
Apple Intelligence: Apple's own on-device AI is limited in what it can do. For anything beyond basic tasks, it falls back to Private Cloud Compute, which means your data goes to Apple's servers. Apple says it's end to end encrypted. You're trusting Apple. And you can't choose your model, load custom models, or use it as a general purpose AI assistant.
Siri with ChatGPT integration: When Siri routes queries to ChatGPT, those queries go to OpenAI. Apple says it strips identifying information, but you're still trusting two companies instead of zero.
All of these services also collect metadata. IP addresses, device information, timestamps, usage patterns.
The Alternative: AI That Never Leaves Your iPhone
Off Grid is a free, open-source app that runs AI entirely on your iPhone's hardware. After you download a model once, the app makes zero network requests. Turn on airplane mode and everything works.
Here's what that actually means in practice:
No account. No Apple ID sign-in, no email, no phone number. Open the app and use it.
No network requests. After the initial model download, Off Grid contacts no server. Ever. Verify it yourself in airplane mode.
No analytics or telemetry. No crash reporting, no usage tracking, no behavioral data collection. The App Store privacy label says it plainly: the developer does not collect any data from this app.
No cloud backup of conversations. Your chats live in the app's local storage on your device. Not in iCloud. Not on a sync server. On your phone.
Open source. MIT licensed. Every line of code is on GitHub. You don't have to trust a privacy policy. You can read the source.
Onboarding
|
Text Generation
|
Image Generation
|
Vision
|
Attachments
|
What You Can Actually Do With It
This isn't a demo or a proof of concept. Off Grid runs six AI capabilities on your iPhone's silicon:
Chat with AI models. Qwen 3, Llama 3.2, Gemma 3, Phi-4, or any GGUF model. Uses Metal GPU acceleration and the Apple Neural Engine. 15 to 30 tokens per second on iPhone 15 Pro and later.
Generate images. On-device Stable Diffusion through Apple's Core ML pipeline. 8 to 15 seconds per image on iPhone 15 Pro. 20+ models. Your prompts and your generated images stay on your phone.
Analyze documents. Attach PDFs, code files, CSVs. Ask questions about them. Nothing gets uploaded. The AI reads them locally on your device.
Voice transcription. On-device Whisper speech to text. Your audio never touches a server. No "recordings may be reviewed to improve our services."
Vision AI. Attach a photo or use your camera and ask questions about what you see. Processed entirely on device.
Tool calling. Models can chain search, calculator, and other tools together in an automatic loop. Even the tool calling logic runs locally.
When Privacy Actually Changes How You Use AI
Most people self-censor with cloud AI without realizing it. You hesitate before typing a health question. You rephrase a personal problem to be vague. You don't upload the actual document because it has client names in it.
When nothing leaves your phone, that hesitation disappears.
Health questions. "What does this symptom mean?" "Is this medication interaction safe?" "What are early signs of [condition]?" On cloud AI, these questions are now part of your permanent profile on someone's server. Locally, they exist only on your phone.
Therapy and journaling. AI is genuinely useful for processing emotions, working through difficult situations, and writing about personal experiences. But only if you're not filtering yourself because you know it's being logged and potentially reviewed by humans.
Legal and financial questions. "What are my rights here?" "How do I handle this debt?" "What does this clause in my lease mean?" Questions that reveal your vulnerabilities and your financial position.
Work and confidential information. Code with proprietary logic. Strategy documents. Client data. If your company's information flows through a cloud AI, it's sitting on a third party's infrastructure.
Creative work. Image generation prompts reveal your creative process and ideas. On Midjourney or DALL-E, every prompt is stored. With Off Grid, your creative process is entirely yours.
The real value of private AI isn't hiding something. It's using the tool fully, without the subconscious filter that kicks in when you know someone might be reading.
How It Works on iPhone
Off Grid downloads models from HuggingFace to your phone's storage. Once downloaded, inference runs on your iPhone's hardware through three compute paths:
Metal (GPU): Handles general purpose parallel computation. This is what llama.cpp uses for text generation.
Apple Neural Engine: A dedicated AI accelerator. Extremely fast and power efficient. Core ML targets the Neural Engine directly for image generation.
CPU: Always available as fallback for smaller models.
The app detects your hardware and uses the fastest path automatically. Models range from 80MB to 4GB+. The model browser filters by your device so you only see what will actually run on your iPhone.
You need an iPhone 12 or newer (A14 chip or later). An iPhone 15 Pro or newer with 8GB of RAM gives you the best experience with larger, more capable models.
Getting Started
- Install Off Grid from the App Store
- Download a recommended model over WiFi
- Turn on airplane mode if you want to verify the privacy claims yourself
- Use AI without anything leaving your phone
Free. Open source. No subscription, no in-app purchases, no ads. Your AI, your device, your data.
Top comments (0)