Surnia – Offline AI Chat

Surnia runs open-source language models directly on your iPhone or iPad. Everything you type stays on the device: there is no account, no server and no tracking.

WHY SURNIA
• Private by design: conversations never leave your device.
• Works offline: once a model is downloaded, you can chat anywhere, with no connection.
• Honest about your device: every model shows whether it suits your iPhone or iPad, from the memory Surnia reads on your device itself.
• Downloads that survive: leave the app and the download keeps going; if it is interrupted, it resumes where it stopped, and it asks before using mobile data.

TALK, LISTEN AND SHARE
• Voice notes: with a model that listens (Gemma 4 on llama.cpp), record up to 30 seconds and the model hears it on your device. The recording is never kept or sent.
• Read aloud: have any reply read to you with your device's own voices.
• Ask Surnia from Siri, Shortcuts or Spotlight.
• Edit a question and send it again, export a chat as Markdown or PDF, and read code with syntax colors.
• With open models, follow-up questions still see a photo you sent earlier.

MODELS
Choose from a curated catalog of open models (Qwen3.5, Gemma 4, Mistral's Ministral, Microsoft's Phi, Liquid LFM2.5, SmolLM3 and ALLaM for Arabic), from tiny and fast to large and capable. On devices with Apple Intelligence, Apple's built-in on-device model is available without any download. You can also import your own GGUF files.

MADE FOR MANY LANGUAGES
Surnia's interface is available in Arabic, Kurdish (Sorani), English, Chinese, Spanish, French, German and Japanese. The models understand many more languages; each model's card lists what it supports. Arabic is fully right-to-left, mixed Arabic and English text reads correctly, and code is always shown left-to-right with a one-tap Copy button.

DESIGNED TO BE SMOOTH
Streaming answers, Markdown formatting, thinking mode for models that support it, attachments (text, code and PDF files; images for vision models), a side panel with your recent chats, and careful memory management so your device stays fast. Share a photo or a file to Surnia from any app with the Share button, and Surnia opens with it, ready for your question.

Surnia uses open-source software including llama.cpp and MLX. Model weights are downloaded from Hugging Face and are subject to their own licenses.