Private AI, on your device.
The only winning move is privacy.
Hopr runs open large language models — and image-generation models — entirely on your iPhone. No cloud, no account, no data collection. Download a quantised open model once, then chat or generate images fully offline, powered by Apple MLX.
How it works
- 01 Pick a model. Choose from a curated library of iPhone-grade chat models, or an image model for text-to-image.
- 02 Download it once. The model streams straight from Hugging Face to your device — the only network request Hopr ever makes.
- 03 Go offline. Every prompt, response, conversation, and generated image stays on your phone. Nothing leaves it. Ever.
On screen
What's inside
-
On-device chat
Live streaming responses, Markdown and code blocks, regenerate, and edit-and-resend.
-
Image generation
Text-to-image with SDXL-Turbo and Stable Diffusion 2.1, rendered inline — tap to zoom, share, regenerate.
-
Model library
Llama 3.2, Qwen2.5 / Qwen3, Gemma 2 / 3, Phi 3.5 — plus any Hugging Face repo via the Experimental section.
-
Conversation history
Search, pin, and delete your chats. Stored on device with SwiftData — one modality per chat.
-
Full generation settings
Temperature, top-p / top-k / min-p, repetition and frequency penalties, max tokens, system prompt, KV-cache tuning.
-
100% private
No account, no telemetry, no profiles. There's nothing to sell because there's nothing collected.
Requirements
- Device
- iPhone 15 or newer (~6 GB+ RAM), iOS 18+
- Engine
- Apple MLX — inference runs on the GPU of a real device
- Download
- Chat models from ~0.7 GB; image models several GB
- Network
- Needed once, to fetch a model. Everything after is offline
- Get it
- Available now on the App Store