On-device AI for iPhone
A private AI that never leaves your phone.
Sunday runs Google's Gemma 4 entirely on your iPhone. Ask it anything, show it a photo, play it a voice note — the model thinks on your device, not on someone's server.
- No account
- No servers
- No data collection
Capabilities
Eleven things it does
without phoning home.
Keep scrolling.
Chat

See
Show it a photo. It tells you what's actually in there.

Hear
Play it a recording and ask what it means.



Watch
Video, sampled frame by frame, then summarised.

Read
Hand it a PDF and ask what you need to do.
Looked it upLive results, only when you say so.
Search
Right hereWeather and time for wherever you are.
Locate
Drop an apple for 30s
0000m
Calculate



Remember
Holds context through long conversations.

Think
Watch it reason through the problem, step by step.
KnowAn on-device knowledge base, for answers with no connection at all.
Live
No upload. No spinner.
It just starts writing.
Nothing leaves the phone, so nothing waits on a round trip. Sunday reads the question and begins answering — in a tunnel, on a plane, or with the radios switched off entirely.
Airplane mode. No Wi-Fi, no cellular, no account.
Privacy
The private version isn't a setting.
It's the architecture.
Most assistants promise not to misuse what you send them. Sunday is built so there's nothing to send.
-
No account, no sign-up
There's no login screen, because there's no server holding a profile of you. Install it and start typing.
-
Your files stay put
Photos, recordings, videos, and PDFs are processed by the model on your device. They are never uploaded for analysis.
-
It works with no connection
The model is on your phone, so it answers in a tunnel, on a plane, or anywhere the signal drops out.
-
The internet is opt-in
Sunday can pull live results for news, scores, and prices — but only when you give it permission, and only at your discretion.
Under the hood
A real frontier model,
quantized to fit.
Google's Gemma 4 runs through MLX, Apple's array framework for Apple silicon — so the neural engine in your phone does the work.
- Model
- Gemma 4 E2B (INT4)
- Engine
- MLX Swift
- Parameters
- ~2.5B effective
- Quantization
- 4-bit
- Context
- 128K tokens
Vision fidelity, video sampling depth, temperature, and response length are all yours to tune. Sunday bundles Gemma 4 weights under Google's Gemma Terms of Use.
Knowledge for the frontier.
Everything runs privately, on your device, at your pace.
Download on theApp Store