Airgap: Private AI Chat

Airgap runs AI models directly on your iPhone — or connects to any model server you control. No account, no relay, no middleman.

Run models on your iPhone
Airgap can now run entirely on device. Use the model built into Apple Intelligence, or download an open model and chat with no network connection at all. On a plane, in a basement, in airplane mode — your prompts never leave the device, because there is nowhere for them to go.
• Apple's built-in model on supported iPhones — nothing to download
• Open models to download: Qwen, Llama, Gemma, Mistral, Phi
• Every model lists its download size and the memory it needs
• Airgap checks what your iPhone can actually run before you download
• Delete any downloaded model to reclaim the space

Or connect to a server you control
Any OpenAI-compatible endpoint — self-hosted (Ollama, LM Studio, vLLM, llama.cpp, LocalAI) or hosted with your own key (OpenRouter, OpenAI, Together, Fireworks, Groq). No accounts, no subscription to any AI provider.

Set up in seconds
1. Choose where your models run — this iPhone, or a server
2. Download a model, or enter your endpoint address
3. Start chatting

Every conversation keeps its own model
Each chat remembers the model it was started with, so switching models in one conversation never disturbs another. Run a small on-device model for quick questions and a 70B on your own hardware for the hard ones.

Built for any AI workflow
• Streaming chat responses
• Markdown rendering with code syntax highlighting
• One-tap copy for messages and code blocks
• Instant model switching, on device or across servers
• Image and file attachments (vision-model dependent)
• Token usage tracking per message
• Thinking and reasoning model support
• Voice input for hands-free prompting
• Automatic compaction so long conversations keep going
• Chat history with local storage
• Encrypted export and import — move your chats between devices

Privacy by architecture
Airgap is not "privacy-focused" as a feature — it's private by design.
• On device, nothing is sent anywhere at all
• No relay — with a server, your device connects to it directly
• No app account or sign-up required
• No message content used for model training
• No analytics on conversation content
• API keys encrypted and stored locally
• Open connection — you see exactly where your data goes

Who Airgap is for
• Anyone who wants a private assistant that works with no signal
• Developers running Ollama or LM Studio on a home machine
• Users connecting to OpenRouter, OpenAI, or Groq without a middleman app
• Privacy-first users who want direct, unproxied AI access
• Home lab and self-hosting enthusiasts
• Teams evaluating private LLM deployments

Why Airgap?
Most AI apps route your prompts through corporate servers. Airgap does the opposite. It runs the model on your iPhone, or talks directly to the endpoint you chose. Whether that's Apple's built-in model, a downloaded open model, a 7B on a Mac mini, or a 70B on a multi-GPU rig, Airgap gives you a polished mobile interface without the middleman.

No middleman. Airgap doesn't relay or store your chats. Just your device and the model you chose.